J. SKELLY WRIGHT, Chief Judge:

This is the third occasion on which this court has been asked to review Environmental Protection Agency (EPA or Agency) regulations promulgated under authority of the Clean Air Act, as amended, 42 U.S.C. § 7401 et seq. (Supp. I 1977) (the Act), and specifically designed to deal with the health problems associated with lead in the ambient air. In Amoco Oil Co. v. EPA, 501 F.2d 722 (D.C.Cir.1974), we upheld regulations prohibiting the. sale of leaded gasoline for use in automobiles equipped with “catalytic converter” devices for controlling exhaust emissions and requiring widespread retail marketing of at least one grade of unleaded gasoline. And in Ethyl Corp. v. EPA, 541 F.2d 1 (D.C.Cir.) (en banc), cert. denied, 426 U.S. 941, 96 S.Ct. 2663, 49 L.Ed.2d 394 (1976), we affirmed an EPA order requiring annual reductions in the lead content of leaded gasoline. In the present consolidated cases we are asked to review EPA regulations establishing national ambient air quality standards for lead. These air quality standards prescribe the maximum concentrations of lead that will be permitted in the air of our country. We must decide whether EPA’s Administrator acted within the scope of his statutory authority in promulgating these regulations and, if so, whether the evidence adduced at the rulemaking proceeding supports his final determinations. In addition, we must examine the petitioners’ claims that infirmities in the procedures employed by EPA in this rulemaking warrant remand of the regulations to the Agency. Petitioners are the Lead Industry Association, Inc. (LIA), a nonprofit trade association whose 78 members include most of the country’s producers and commercial consumers of lead (No. 78-2201), and St. Joe Minerals Corporation (St. Joe) (No. 78-2220).1

I. BACKGROUND

Man’s ability to alter his environment to achieve perceived goals has undoubtedly made an enormous contribution to his economic and social well-being. This undertaking is not, however, without attendant costs. One of these costs is the toll that these alterations may exact on the environment itself and, in turn, the dangers that this may pose for the public health and welfare. Unfortunately, man’s ability to alter the environment often far outstrips his ability to foresee with any degree of certainty what untoward effects these changes may bring. The issues presented by these cases illustrate this sad fact.

Lead’s environmental significance is a consequence of both its abundance and its utility. The relative abundance of lead in the earth’s crust makes it unique among the toxic heavy metals. EPA’s “Air Quality Criteria For Lead” (hereinafter cited as CD) 1-1, Joint Appendix (JA) 1105. And centuries of mining and smelting, and the use of lead in a variety of human activities, have increased the natural background concentration of lead in the environment. Id. But it is only since the industrial age and the use of lead as a gasoline additive that lead has become pervasive. Id. at 1-2 — 1-3, JA 1106-1107. Today lead is ubiquitous. It is found in almost every medium with which we come into contact — food, water, air, soil, dust, and paint, id. 1-1, JA 1105, each of which represents a potential pathway for human lead exposure through ingestion or inhalation. The widespread presence of this toxic metal in the environment poses a significant health risk. Lead is a poison which has no known beneficial function in the body, id. 1-12, JA 1116, but when present in the body in sufficient concentrations lead attacks the blood, kidneys, and central nervous arid other systems arid can cause anemia, kidney damage, severe *1136brain damage, and death. Id. 1-6 — 1-9, JA 1110-1113.2

There are three major sources of the body’s lead burden. In most people the largest source is diet. CD 7-9, JA 1179.3 Another source, particularly in children, is the habit of placing hands, objects, and materials in the mouth.4 The third major source is the ambient air; airborne lead is deposited in the respiratory tract as a person breathes lead-contaminated air and is subsequently absorbed into the bloodstream. CD 1-5, JA 1108. Once the lead is in the bloodstream its source is immaterial; total lead intake is the sum of the intake from all these sources. The multiplicity of sources of lead intake increases the difficulty of controlling human lead exposure. Much of the protective activity in this area has focused on limiting the amount of lead in the ambient air, the most controllable source of lead exposure. In this country, by far the largest source of lead emissions— accounting for 88 percent of total lead emissions according to EPA estimates — is the exhaust of motor vehicles powered by gasoline containing lead additives. CD 5-3, JA 1140. Another eight percent of lead emissions is the result of solid waste incineration and combustion of waste oil. Id. Industrial facilities account for the remaining four percent of total lead emissions. Id.

Acting pursuant to authority conferred on it by Congress in the Clean Air Act, as amended, 42 U.S.C. § 7401 et seq., EPA has been involved in regulation of lead emissions almost since the Agency’s inception.5 Its initial approach to controlling the amount of lead in the ambient air was to limit lead emissions from automobiles by restricting the amount of lead in gasoline. To this end it promulgated the regulations which we upheld in Amoco Oil Corp. v. EPA, supra, and Ethyl Corp. v. EPA, supra. However, in 1975 the Natural Resources Defense Council, Inc. (NRDC), and others brought suit against EPA claiming that the Agency was required by Section 108 of the Clean Air Act, 42 U.S.C. § 7408, to list lead as a pollutant for which an air quality criteria document would be prepared, and for which national ambient air quality standards should be promulgated under Section 109 of the Act, 42 U.S.C. § 7409. The District Court agreed with NRDC and directed the Administrator to list lead as a pollutant under Section 108 of the Act, by March 31, 1976. Natural Resources Defense Council, Inc. v. Train, 411 F.Supp. 864 (S.D.N.Y.1976). The Second Circuit affirmed, 545 F.2d 320 (2d Cir. 1976), and EPA initiated the proceedings outlined in the statute which are under review here.

II. THE STATUTORY SCHEME

The first step toward establishing national ambient air quality standards for a particular pollutant is its addition to a list, compiled by EPA’s Administrator, of pollutants that cause or contribute to air pollution “which may reasonably be anticipated to endanger public health or welfare[.]” Section 108(a)(1), 42 U.S.C. § 7408(a)(1). Within twelve months of the listing of a pollutant under Section 108(a) the Administrator must issue “air quality criteria” for the pollutant. Section 108 makes it clear that the term “air quality criteria” means something different from the conventional meaning of “criterion”; such “criteria” do not constitute “standards” or “guidelines,” but rather refer to a document to be prepared by EPA which is to provide the scien-*1137tifie basis for promulgation of air quality standards for the pollutant. This criteria document must “accurately reflect the latest scientific knowledge useful in indicating the kind and extent of all identifiable effects on public health or welfare which may be expected from the presence of such pollutant in the ambient air, in varying quantities.” Section 108(a)(2), 42 U.S.C. § 7408(a)(2).

At the same time as he issues air quality criteria for a pollutant, the Administrator must also publish proposed national primary and secondary air quality standards for the pollutant. Section 109(a)(2), 42 U.S.C. § 7409(a)(2). National primary ambient air quality standards are standards “the attainment and maintenance of which in the judgment of the Administrator, based on such criteria and allowing an adequate margin of safety, are requisite to protect the public health.” Section 109(b)(1), 42 U.S.C. § 7409(b)(1). Secondary air quality standards “specify a level of air quality the attainment and maintenance of which in the judgment of the Administrator, based on such criteria, is requisite to protect the public welfare from any known or anticipated adverse effects associated with the presence of such air pollutant in the ambient air.” Section 109(b)(2), 42 U.S.C. § 7409(b)(2). Effects on “the public welfare” include “effects on soils, water, crops, vegetation, manmade materials, animals, wildlife, weather, visibility, and climate, damage to and deterioration of property, and hazards to transportation, as well as effects on economic values and on personal comfort and well-being.” Section 302(h), 42 U.S.C. § 7602(h). The Administrator is required to submit the proposed air quality standards for public comment in a rulemak-ing proceeding, the procedure for which is prescribed by Section 307(d) of the Act, 42 U.S.C. § 7607(d).

Within six months of publication of the proposed standards the Administrator must promulgate final primary and secondary ambient air quality standards for the pollutant. Section 307(d)(10), 42 U.S.C. § 7607(d)(10). Once EPA has promulgated national ambient air quality standards, responsibility under the Act shifts from the federal government to the states. Within nine months of promulgation of the standards each state must prepare and submit to EPA for approval a state implementation plan. Section 110(a)(1), 42 U.S.C. § 7410(a)(1). These state implementation plans must contain emission limitations and all other measures necessary to attain the primary standards “as expeditiously as practicable,” but no later than three years after EPA approval of the plan, and to attain the secondary standards within a reasonable period of time. Section 110(a)(2)(A) & (B), 42 U.S.C. § 7410(a)(2)(A) & (B). The Administrator is authorized to extend the deadline for attainment of the primary air quality standards by two years, but thereafter it must be met. Section 110(e), 42 U.S.C. § 7410(e).

III. THE LEAD STANDARDS RULE-MAKING PROCEEDINGS

As required by statute, EPA’s first step toward promulgating air quality standards for lead was to prepare a criteria document. The Lead Criteria Document was the culmination of a process of rigorous scientific and public review, and thus is a comprehensive and thoughtful analysis of the most current scientific information on the subject. The Lead Criteria Document went through three major drafts, and three separate reviews, including public meetings by the Subcommittee on Scientific Criteria for Environmental Lead of EPA’s Science Advisory Board (SAB Lead Subcommittee).6 The Agency reviewed over 280 public comments, most of a sophisticated scientific nature, before it issued the final Criteria Document. Members of the public, industry (in-*1138eluding the petitioners in these cases), environmental groups, the scientific community, and state and federal government agencies actively participated in the review of the drafts. Notice of the meetings of the SAB Lead Subcommittee was published in the Federal Register, and the drafts of the Criteria Document which were to be reviewed were available before the meetings. A formal record and a transcript of the proceedings were kept, and a review of the transcript shows that scientists with differing views could — and did — exchange ideas with each other as well as agency staff, and that all were questioned by the members of the Subcommittee.

A. The Lead Criteria Document7

EPA released its “Air Quality Criteria For Lead” on December 14, 1977. 42 Fed. Reg. 63076, JA 1480. The document was “prepared to reflect the current state of knowledge about lead — specifically, those issues that are most relevant to establishing the objective scientific data base that will be used to recommend an air quality standard for lead that will adequately safeguard the public health.” CD 1-1, JA 1105. Accordingly, the Criteria Document examined a large number of issues raised by the problem of lead in the environment. One of these was the effects of lead exposure on human health. The Criteria Document concluded that, among the major organ systems, the hematopoietic (blood-forming) and neurological systems are the areas of prime concern. CD 1-12, JA 1116. Its discussion of the effects of lead on these two organ systems is central to our review of the lead standards.8

The Criteria Document identified a variety of effects of lead exposure on the blood-forming system. We will discuss only the effects that played an important role in the Administrator’s analysis. Anemia, which can be caused by lead-induced deformation and destruction of erythrocytes (red blood cells) and decreased hemoglobin synthesis,9 is often the earliest clinical manifestation of lead intoxication. CD 11-7,11-8,11-13, JA 1229-1230, 1235. Symptoms of anemia include pallor of the skin, shortness of breath, palpitations of the heart, and fatigability.10 The Criteria Document concluded, after a review of various studies, that in “children, a threshold level for anemia is about 40 u[g] *1139Pb/dl, whereas the corresponding value for adults is about 50 ug Pb/dl.” CD 11-13, JA 1235. (The concentration of lead in the blood is measured in micrograms of lead per deciliter of blood — ug Pb/dl.)

The Criteria Document also examined other more subtle effects on the blood-forming system, associated with lower levels of lead exposure. The most pertinent of these “subclinical”11 effects for purposes of these cases is lead-related elevation of erythrocyte protoporphyrin (EP elevation).12 According to the Criteria Document, this phenomenon must, for a number of reasons, be regarded as an indication of an impairment of human health. CD 11 — 11, JA 1233. First, EP elevation indicates an impairment in the functioning of the mitochondria, the subcellular units which play a crucial role in the production of energy in the body, and in cellular respiration. CD 1-6, 11-11, 11-14, 13-5, JA 1110, 1233, 1236, 1335. Second, it indicates that lead exposure has begun to affect one of the basic biological functions of the body — production of heme within the red blood cells.13 Heme is critical to transporting oxygen to every cell in the body. Third, EP elevation may indicate that any reserve capacity there may be in the heme synthesis system has been reduced. CD 11-11, JA 1233. Finally, the Criteria Document noted that lead’s interference with the process of heme synthesis in the blood may suggest that lead interferes with production of heme proteins in other organ systems, particularly the renal and neurological systems. CD 11-11, JA 1233. The Criteria Document reported that the threshold for EP elevation in children and women is at blood lead levels of 15-20 ug Pb/dl, and 25-30 ug Pb/dl in adult males. CD 13-8 (Table 13-2), JA 1338 (Table 13-2). While suggesting that some of the initial hematological effects of lead exposure may constitute relatively mild effects at low blood lead levels, the Criteria Document concluded that “they nevertheless signal the onset of steadily intensifying adverse effects as blood lead elevations increase. Eventually, [these] * * * effects reach such magnitude that they are of clearcut medical significance as indicators of undue lead exposure.” CD 1-13, JA 1117. The Criteria Document did not identify a particular blood lead level at which regulatory response was appropriate, but it did note with approval the 1975 guidelines issued by the Center For Disease Control, which use elevated EP at blood lead levels of 30 ug Pb/dl as the cut-off point in screening children for lead poisoning. CD 13-5, JA 1335.

The Criteria Document also examined the effects of lead exposure on the central nervous system. Among the most deleterious effects of lead poisoning are those associated with severe central nervous system *1140damage at high exposure levels. The Criteria Document noted that neurological and behavioral deficits have long been known to be among the more serious effects of lead exposure, but it pointed out that there is disagreement about whether these effects are reversible, and about what exposure levels are necessary to produce specific deleterious effects. CD 11-14, JA 1236. Much of the impetus for the debate on these questions has been provided by the continual emergence of new information suggesting that lead exposure levels previously thought to be harmless actually cause significant neurological damage. Id. The more severe neurological effects of high level lead exposure are the clinical syndrome of lead encephalopathy. Early symptoms include dullness, restlessness, irritability, headaches, muscular tremor, hallucinations, and loss of memory. These symptoms rapidly progress (sometimes within 48 hours) to delirium, mania, convulsions, paralysis, coma, and death. Id. at 11-15, JA 1237. The Criteria Document expressed particular concern that the onset of these serious symptoms can be quite abrupt, even in the absence of prior overt or clinical symptoms of disease. Id. at 13-6, JA 1336. After a review of various studies, the Criteria Document concluded that the blood lead threshold for these neurological effects of high level exposure is 80-100 ug Pg/dl in children, and 100-200 ug Pb/dl in adults. Id. at 1-13, 11-18, 11-25, 13-6, JA 1117, 1240, 1247, 1336.

The Criteria Document also went on to consider the evidence on whether lower level lead exposures can affect the central nervous system, particularly in children. It acknowledged that the issue is unsettled and somewhat controversial, but it was able to conclude, after a careful review of various studies on the subject,14 that “a rather consistent pattern of impaired neural and cognitive functions appears to be associated with blood lead levels below those producing the overt symptomatology of lead encephalopathy.” CD 1-7, JA 1111. The Criteria Document reported that “[t]he blood lead levels at which neurobehavioral deficits occur in otherwise asymptomatic children appear to start at a range of 50 to 60 ug/dl, although some evidence tentatively suggests that such effects may occur at slightly lower levels for some children.” Id.15

In addition to examining the health effects of lead exposure, the Criteria Document also discussed other issues critical to the task of setting air quality standards for lead. One of these issues is the relationship between air lead exposure and blood lead levels — a relationship commonly referred to as the air lead/blood lead ratio. The Criteria Document acknowledged that derivation of a functional relationship between air lead exposure and blood lead levels is made difficult by the fact that the relationship is not a linear one; rather, the ratio tends to increase as air lead levels are reduced, CD 12-24, JA 1311. The Document was nevertheless able to conclude, after a detailed examination of the relevant studies, CD 12-22 — 12-29, JA 1309-1316, that air lead/blood lead ratios fall within a range of 1:1 to 1:2 (ug Pb/m3 air):(ug Pb/dl blood) at the levels of lead exposure generally encountered by the population, i.e., blood lead levels increase by between 1 and 2 ug Pb/dl of blood for every 1 ug Pb/m3 of air. (Air lead content is measured in micrograms of lead per cubic meter of air — ug Pb/m3.) CD 12-38, JA 1325. The Criteria Document *1141reported that the studies indicate that the ratio for children is at the upper end of this range or even slightly above it. Id.

Finally, the Criteria Document also examined the distribution of blood lead levels throughout the population, concluding that there is a significant variability in individual blood lead responses to any particular level of air lead exposure. It further found that this variability is consistent and predictable, and that the application of established statistical techniques to the distribution of individual blood lead levels would make it possible to predict what proportion of the population would be above or below any particular blood lead level at a given level of air lead exposure.16 The Criteria Document looked into the question whether any sub-groups within the population are particularly vulnerable to the effects of lead exposure. It concluded that preschool-age children and pregnant women are particularly sensitive to lead exposure, the latter mainly because of the risk to the unborn child. CD'13-11 — 13-14, JA 1341-1344.

B. The Proposed Standards

Simultaneously with the publication of the Lead Criteria Document on December 14, 1977, the Administrator proposed a national primary ambient air quality standard for lead of 1.5 ug Pb/m3 monthly average. 42 Fed.Reg. 63076, JA 1480. He also proposed that the secondary air quality standard be set at the same level as the primary standard because the welfare effects associated with lead exposure did not warrant imposition of a stricter standard. 42 Fed. Reg. 63081-63082, JA 1485-1486. In the preamble to the proposed standards the Administrator explained the analysis EPA had employed in setting the standards.

The Administrator first pointed out that a number of factors complicate the task of setting air quality standards which will protect the population from the adverse health effects of lead exposure. First, some subgroups within the population have a greater potential for, or are more susceptible to the effects of, lead exposure. Id. at 63077, JA 1481. Second, there are a variety of adverse health effects associated with various levels of lead exposure. Id. Third, the variability of individual responses to lead exposure, even within particular sub-groups of the population, would produce a range of blood lead levels at any given air lead level. Id. at 63079, JA 1483. Fourth, airborne lead is only, one of a number of sources of lead exposure and the relative contribution from each source is difficult to quantify. Id. at 63080, JA 1484. Finally, the relationship between air lead exposure and blood lead levels is a complex one. Id. at 63079, JÁ 1483.

In response to the first problem the Administrator began by noting that protection of the most sensitive groups within the population had to be a major consideration in determining the level at which the air quality standards should be set. And he determined that children between the ages of 1 and 5 years are most sensitive to the effects of lead exposure both because the hematologic and neurologic effects associated with lead exposure occur in children at lower threshold levels than in adults, and because the habit of placing hands and other objects in the mouth subjects thém to a greater risk of exposure. Id. at 63077-63078, JA 1481-1482. Next, the Administrator examined the various health effects of lead exposure and proposed that EP elevation should be considered' the first adverse health effect of lead exposure because it indicates an impairment of cellular functions, and should be the pivotal health effect on which the lead standards are based. Id. at 63078, JA 1482. Accordingly, he proposed that the air lead standards be designed to prevent the occurrence of EP elevation in children. In order to accom*1142plish this, and to address the problem of variable responses to lead exposure, the Administrator selected 15 ug Pb/dl, the lowest reported threshold blood lead level for EP elevation in children, as the target mean population blood lead level.17 He reasoned that setting the target mean population blood lead level at the lowest reported threshold blood lead level for EP elevation would ensure that most of the target population would be kept below blood lead levels at which adverse health effects occur. Id. at 63078, JA 1483. The Administrator also discussed the alternative approaches of basing the standard on more severe effects such as anemia, or attempting to decide the actual level of EP elevation which represents an adverse effect on health, and then making an adjustment to allow a margin of safety. Id. He specifically invited comments on these alternative approaches. Id. Finally, the Administrator outlined another approach to calculating the target mean population blood lead level involving the use of statistical techniques discussed in the Criteria Document. Id18

Having selected a target mean population blood lead level, the Administrator’s next step was to allow for the multiplicity of sources of lead exposure. He thus had to estimate the amount of blood lead that should be attributed to non-air sources. The Administrator admitted that any amount he selected could be no more than a theoretical national average, and on the basis of the evidence available he proposed that the lead standards should be based on the general assumption that 12 ug Pb/dl of blood lead should be attributed to non-air sources. Id. at 63080-63081, JA 1484-1485. Given the target mean population blood lead level of 15 ug Pb/dl and the assumed contribution from non-air sources of 12 ug Pb/dl, the maximum allowable contribution from ambient air is 3 ug Pb/dl. The final step in his analysis was to determine what air lead level would prevent the ambient air contribution to blood lead levels from exceeding 3 ug Pb/dl. This step required determining the relationship between air lead exposure and blood lead levels, i.e., the air lead/blood lead ratio. On the basis of the information in the Criteria Document, the Administrator selected a ratio of 1:2 as appropriate for calculating the effect of air lead exposure on blood lead levels in children. Id. at 63079, JA 1483.

Thereafter, calculation of the air quality standard was a mathematical exercise as shown in the following table.

The Administrator concluded, on the basis of available information, that the averaging period for the lead standard should be a calendar month. Id. at .63081, JA 1485.

C. Public Comments

The public comment period ran from December 14, 1977 to March 17, 1978, and public hearings on the proposed standards were held on February 15 and 16, 1978. 43 Fed.Reg. 46246, JA 2948. The comments on *1143the proposed standards were sharply divided. The comments submitted by the lead industry and its experts uniformly opposed the proposed standards, and many endorsed a standard of 5 ug Pb/m3, the standard proposed in the discredited first draft of the Criteria Document, see note 7 supra, as adequate to protect the public health. 43 Fed.Reg. 46248, JA 2950. On the other hand, environmental groups, medical experts, and state, local, and federal agencies either endorsed the proposed standards or called for even stricter standards. Id. None of the comments seriously questioned the selection of children between the ages of 1 and 5 years as the target population group, or the estimate of a contribution from non-air sources of 12 ug Pb/dl. The major areas of controversy were the Administrator’s choice of EP elevation as the pivotal adverse health effect and his conclusion that the threshold blood lead level for EP elevation in children is 15 ug Pb/dl, the selection of an appropriate air lead/blood lead ratio, the appropriate allowance for an adequate margin of safety, and the averaging time period for the standards. Id.19

A number of comments challenged the selection of EP elevation as the pivotal adverse health effect, insisting that EP elevation merely indicates a biological change or response which is in no way harmful to health,20 and in addition they criticized the Administrator’s determination that the blood lead threshold for EP elevation in children is 15 ug Pb/dl.21 These comments suggested that a decrease in hemoglobin levels, which begins at blood lead levels no lower than 40 ug Pb/dl, should be the pivotal adverse health effect on which the standards are based.22 Other experts, however, agreed with the Administrator’s conclusion that EP elevation must be considered an adverse health effect of lead exposure, and argued that using EP elevation as the pivotal adverse health effect would, in addition, allow an adequate margin of safety in protecting against the more serious health effects associated with higher levels of lead exposure.23 Finally, several industry experts appeared to indicate a preference for the lognormal statistical procedures that the Administrator had, in the proposed standards, suggested as an alternative method for determining the target mean population blood lead level.24

D. The Final Air Quality Standards for Lead

The Administrator promulgated the final air quality standards on October 5, 1978, prescribing national primary and secondary ambient air quality standards for lead of 1.5 ug Pb/m3, averaged over a calendar quarter. 43 Fed.Reg. 46246, JA 2948. Although the final standards were the same as the proposed standards (with the exception of the change in the averaging period from 30 to 90 days), the Administrator arrived at the final standards through somewhat different analysis. The preamble to the final standards reveals that the comments on the proposed standards had led the Administrator to reconsider his analysis. In particular, he seemed to feel that legitimate questions *1144had been raised concerning the health significance of the early stages of EP elevation and about the threshold blood lead level for this condition. 43 Fed.Reg. 46248, 46253, JA 2950, 2955. The Administrator’s reexamination focused on two key questions: (1) What is the maximum safe individual blood lead level for children? and (2) what proportion of the target population should be kept below this blood lead level? Id. at 46249, 46252-46253, JA 2951, 2954-2955. Addressing the first issue required a review of the health effects of lead exposure discussed in the Criteria Document. . The Administrator concluded that, although EP elevation beginning at blood lead levels of 15-20 ug Pb/dl is potentially adverse to the health of children, only when blood lead concentration reaches a level of 30 ug Pb/dl is this effect significant enough to be considered adverse to health. Id. at 46253, JA 2955. Accordingly, he selected 30 ug Pb/dl as the maximum safe individual blood lead level for children. Id. The Administrator based this choice on three mutually supporting grounds. First, it is at this blood lead level that the first adverse health effect of lead exposure — impairment of heme synthesis — begins to occur in children. Second, a maximum safe individual blood lead level of 30 ug Pb/dl would allow an adequate margin of safety in protecting children against more serious effects of lead exposure — anemia, symptoms of which begin to appear in children at blood lead levels of 40 ug Pb/dl, and central nervous system deficits which start to occur in children at blood lead levels of 50 ug Pb/dl. Third, the Administrator reasoned that the maximum safe individual blood lead level should be no higher than the blood lead level used by the Center for Disease Control in screening children for lead poisoning — 30 ug Pb/dl. Id.

Having determined the maximum safe individual blood lead level for the target population, the Administrator next focused on the question of what percentage of children between the ages of 1 and 5 years the standard should attempt to keep below this blood lead level. According to the 1970 census, there are approximately 20 million children under the age of 5 years in the United States, 12 million of them in urban areas and 5 million in inner cities where lead exposure may be especially high. The Administrator concluded that in order to provide an adequate margin of safety, and to protect special high risk subrgroups, the standards should aim at keeping 99.5% of the target population below the maximum safe individual blood lead level of 30 ug Pb/dl.25 Id. at 46253, 46255, JA 2955, 2957. The next step in the analysis was to determine what target mean population blood lead level would ensure that 99.5% of the children below the age of 5 years would be kept below the maximum safe individual blood lead level of 30 ug Pb/dl. Using the lognormal statistical technique he had alluded to in the proposed standards,26 he calculated that a target mean population blood lead level of 15 ug Pb/dl (the same number as in the proposed standards, but arrived at through different analysis), would accomplish this task.27 Id. at 46253, 46254, JA 2955, 2956. Thereafter, the Administrator used the same estimate of the contribution from non-air sources, 12 ug Pb/dl, and the same air lead/blood lead ratio, 1:2, that he had used in calculating the proposed standards,28 to compute the final ambient air quality standards for lead. The result was an ambient air quality stan*1145dard of 1.5 ug Pb/m3, the same as the proposed standard. Id. at 46254, JA 2956. The Administrator did, however, change the averaging period for the standards from one calendar month to one calendar quarter, id. at 46255, JA 2957, because he felt that this change would significantly improve the validity of the data to be used in monitoring the progress toward attainment of the standards without rendering the standards less protective. Id.

On December 8,1978 LIA petitioned EPA for reconsideration and a stay of the lead standards. JA 2980-3000. The Administrator denied the petition on February 2, 1979. JA 3001-3007. These petitions for review of the lead standards regulations followed. Before examining the petitioners’ challenges to the regulations, we consider the limits of our reviewing function.

IV. STANDARD OF REVIEW

The scope of judicial review of the Administrator’s decisions and actions is delineated by Section 307(d) of the Act, 42 U.S.C. § 7607(d). We must uphold the Administrator’s actions unless we find that they were: (1) “arbitrary, capricious, an abuse of discretion, or otherwise not in accordance with law”; (2) “contrary to constitutional right, power, privilege, or immunity”; (3) “in excess of statutory jurisdiction, authority, or limitations, or short of statutory right[.]” Section 307(d)(9), 42 U.S.C. § 7607(d)(9). In addition, we may set aside any action found to be “without observance of procedure required by law,” if (i) the failure to follow the prescribed procedure was arbitrary or capricious, (ii) the procedural objection was raised during the public comment period, or there were good reasons why it was not, and (iii) the procedural errors “were so serious and related to matters of such central relevance to the rule that there is a substantial likelihood that the rule would have been significantly changed if such errors had not been made.” Id. Section 307(d)(8), 42 U.S.C. § 7607(d)(8).

These statutory provisions and a considerable body of ease law demonstrate that our role as a reviewing court is limited. The “arbitrary and capricious” standard of review is highly deferential, and presumes agency action to be valid. Citizens to Preserve Overton Park, Inc. v. Volpe, 401 U.S. 402, 415, 91 S.Ct. 814, 823, 28 L.Ed.2d 136 (1971); Ethyl Corp. v. EPA, supra, 541 F.2d at 34. Moreover, the reviewing court may not substitute its judgment for the agency’s, Citizens to Preserve Overton Park, Inc. v. Volpe, supra, 401 U.S. at 416, 91 S.Ct. at 823, and must affirm the agency’s decision if a rational basis for it is presented. Bowman Transportation, Inc. v. Arkansas-Best Freight Systems, Inc., 419 U.S. 281, 290, 95 S.Ct. 438, 444, 42 L.Ed.2d 447 (1974); United States v. Allegheny-Ludlum Steel Corp., 406 U.S. 742, 749, 92 S.Ct. 1941, 1946, 32 L.Ed.2d 453 (1972). Of course a reviewing court does not serve as a mere rubber stamp for agency decisions. Rather, the function of judicial review is to ensure that agency decisions are “based on a consideration of the relevant factors.” Citizens to Preserve Overton Park, Inc. v. Volpe, supra, 401 U.S. at 416, 91 S.Ct. at 824.

In addition, the court must undertake a “substantial inquiry” into the facts, one that is “searching and careful.” Id. at 415, 416, 91 S.Ct. at 823; Ethyl Corp. v. EPA, supra, 541 F.2d at 34. In cases such as the ones we have before us, cases which involve complex scientific and technical questions, conducting a “substantial inquiry” into the facts may require the court to delve into the scientific literature. The purpose of this scrutiny of the evidence in the record is to educate the court. As we pointed out in Ethyl Corp. v. EPA, supra, the court “must understand enough about the problem confronting the agency to comprehend the meaning of the evidence relied upon and the evidence discarded; the questions addressed by the agency and those bypassed; • the choices open to the agency and those made.” 541 F.2d at 36. Only by doing this can the court “penetrate to the underlying decisions of the agency, to satisfy itself that the agency has exercised a reasoned discretion, with reasons that do not deviate from or ignore the ascertainable legislative intent.” Greater Boston Televi*1146sion Corp. v. FCC, 444 F.2d 841, 850 (D.C. Cir.1970), cert. denied, 403 U.S. 923, 91 S.Ct. 2229, 29 L.Ed.2d 701 (1971).29 However, it is appropriate to sound some notes of caution about the limits of this exercise. First, we would be less than candid if we failed to acknowledge that we approach the task of examining some of the complex scientific issues presented in cases of this sort with some diffidence.30 More important, we stress that our review of the evidence is not designed to enable us to second-guess the Agency’s expert decisionmaker. Ethyl Corp. v. EPA, supra, 541 F.2d at 36. Congress has entrusted the Agency with the responsibility for making these scientific and other judgments, and we must respect both Congress’ decision and the Agency’s ability to rely on the expertise that it develops. Market Street Railway v. Railroad Commission, 324 U.S. 548, 559-561, 65 S.Ct. 770, 776-77, 89 L.Ed. 1171 (1945); Ethyl Corp. v. EPA, supra, 541 F.2d at 36. As we explained in Ethyl Corp. v. EPA :

[Ajfter our careful study of the record, we must take a step back from the agency’s decision. We must look at the decision not as the chemist, biologist or statistician that we are qualified neither by training nor experience to be, but as a reviewing court exercising our narrowly defined duty of holding agencies to certain minimal standards of rationality. “Although [our] inquiry into the facts is to be searching and careful, the ultimate standard of review is a narrow one.” * * We must affirm unless the agency decision is arbitrary or capricious.

541 F.2d at 36-37 (citations and footnotes omitted; brackets in original).

It is also important to note that although the pertinent sections of the Clean Air Act outline the policy objectives to be sought and the procedural framework to be followed in promulgating ambient air quality standards, Congress left the formulation of the specific standards to EPA’s Administrator. This task presents complex questions of science, law, and social policy under the Act. The record is lengthy — approximately 10,000 pages — and it is highly technical. The Administrator’s task required both “a legislative policy determination and an adjudicative resolution of disputed facts.” Mobil Oil Corp. v. FPC, 483 F.2d 1238, 1257 (D.C.Cir.1973).

These are conceptually distinct types of decisions, and it is important that we keep this in mind in reviewing the Administrator’s decisions. See Industrial Union Dep’t, AFL-CIO v. Hodgson, 499 F.2d 467, 474-475 (D.C.Cir.1974). Where factual determinations were necessary the Administrator often had to make decisions in the face of conflicting evidence. In some instances this merely required that he draw conclusions from the evidence in the record. In reviewing these conclusions we can examine the record to ascertain whether there is substantial evidence in the record when considered as a whole which supports the Administrator’s determinations. Id. at 474. Other questions involved in the standard-*1147setting process, however, are at the very “frontiers of scientific knowledge.”31 Consequently, the information available may be insufficient to permit fully informed factual determinations. In such instances the Administrator’s decisions necessarily had to rest largely on policy judgments. Policy choices of this sort “are not susceptible to the same type of verification or refutation by reference to the record as are [other] factual questions.” 499 F.2d at 475. While we will indeed scrutinize such judgments carefully, we must adopt a different mode of judicial review. Id.; accord, Nat’l Asphalt Pavement Ass’n v. Train, 539 F.2d 775, 783-784 (D.C.Cir.1976); Automotive Parts & Accessories Ass’n, Inc. v. Boyd, 407 F.2d 330, 336, 338 (D.C.Cir.1968). As this court has previously stated:

What we are entitled to * * * is a careful identification by the [Administrator], when his proposed standards are challenged, of the reasons why he chooses to follow one course rather than another. Where that choice purports to be based on certain determinable facts, [he] must, in form as well as substance, find those facts from evidence in the record. By the same token, when the [Administrator] is obliged to make policy judgments where no factual certainties exist or where facts alone do not provide the answer, he should so state and go on to identify the considerations he found persuasive.”

Industrial Union Dep’t, AFL-CIO v. Hodg-son, supra, 499 F.2d at 475-476. In short, “[t]he paramount objective is to see whether the agency, given an essentially legislative task to perform, has carried it out in a manner calculated to negate the dangers of arbitrariness and irrationality in the formulation of rules for general application in the future.” Automotive Parts & Accessories Ass’n, Inc. v. Boyd, supra, 407 F.2d at 338.

Finally, although we may set aside the Administrator’s decisions if we find that he exceeded his authority under the statute, we note that EPA’s construction of the Clean Air Act has been accorded considerable deference by the courts. Union Electric Co. v. EPA, 427 U.S. 246, 256, 96 S.Ct. 2518, 2525, 49 L.Ed.2d 474 (1976); Train v. Natural Resources Defense Council, Inc., 421 U.S. 60, 75, 95 S.Ct. 1470, 1479, 43 L.Ed.2d 731 (1975); Ethyl Corp. v. EPA, supra, 541 F.2d at 12 n.16.32 Where different interpretations of the statute are plausible, so - long as EPA’s construction of the statute is reasonable we may not substitute our own interpretation for the Agency’s. Train v. Natural Resources Defense Council, Inc., supra, 421 U.S. at 75, 95 S.Ct. at 147933 “[T]he construction of a statute by those charged with its execution should be followed unless there are compelling indications that it is wrong[.]” Red Lion Broadcasting Co. v. FCC, 395 U.S. 367, 381, 89 S.Ct. 1794, 1802, 23 L.Ed.2d 371 (1967); accord, Beal v. Doe, 432 U.S. 438, 97 S.Ct. 2366, 53 L.Ed.2d 464 (1977). Deference to the Administrator’s interpretation is particularly appropriate in construing a statute that invests him with a considerable amount of discretion. Unless it can be shown that the Administrator’s construction of the statute is plainly unreasonable, we must uphold his interpretation. Motor & Equipment Manufacturers Ass’n, Inc. v. EPA, 627 F.2d 1095, at 1106 (D.C.Cir.1979).

Thus mindful of our restricted role, we turn to consider petitioners’ claims. Petitioners posit three basic questions for deci*1148sion. First, did the Administrator exceed his authority under the statute in promulgating the lead standards? Second, were key elements in the Administrator’s analysis arbitrary or capricious? Third, do alleged procedural shortcomings in the lead standards rulemaking warrant a remand of the regulations to EPA?

V. STATUTORY AUTHORITY

The petitioners’ first claim is that the Administrator exceeded his authority under the statute by promulgating a primary air quality standard for lead which is more stringent than is necessary to protect the public health because it is designed to protect the public against “sub-clinical” effects which are not harmful to health. According to petitioners, Congress only authorized the Administrator to set primary air quality standards that are aimed at protecting the public against health effects which are known to be clearly harmful. They argue that Congress so limited the Administrator’s authority because it was concerned that excessively stringent air quality standards could cause massive economic dislocation.

In developing this argument St. Joe contends that EPA erred by refusing to consider the issues of economic and technological feasibility in setting the air quality standards for lead. St. Joe’s claim that the Administrator should have considered these issues is based on the statutory provision directing him to allow an “adequate margin of safety” in setting primary air quality standards. In St. Joe’s view, the Administrator must consider the economic impact of the proposed standard on industry and the technological feasibility of compliance by emission sources in determining the appropriate allowance for a margin of safety.34 St. Joe argues that the Administrator abused his discretion by refusing to consider these factors in determining the appropriate margin of safety for the lead standards, and maintains that the lead air quality standards will have a disastrous economic impact on industrial sources of lead emissions.

This argument is totally without merit. St. Joe is unable to point to anything in either the language of the Act or its legislative history that offers any support for its claim that Congress, by specifying that the Administrator is to allow an “adequate margin of safety” in setting primary air quality standards, thereby required the Administrator to consider economic or technological feasibility. To the contrary, the statute and its legislative history make clear that economic considerations play no part in the promulgation of ambient air quality standards under Section 109.

Where Congress intended the Administrator to be concerned about economic and technological feasibility, it expressly so provided. For example, Section 111 of the Act, 42 U.S.C. § 7411, directs the Administrator to consider economic and technological feasibility in establishing standards of performance for new stationary sources of air pollution based on the best available control technology. See Nat’l Asphalt Pavement Ass’n v. Train, supra, 539 F.2d 775; S.Rep. No.91-1196, 91st Cong., 2d Sess. 416 (1970).35 In contrast, Section 109(b) speaks only of protecting the public health and welfare.36 Nothing in its language suggests that the Administrator is to consider *1149economic or technological feasibility in setting ambient air quality standards.37

The legislative history of the Act also shows the Administrator may not consider economic and technological feasibility in setting air quality standards; the absence of any provision requiring consideration of these factors was no accident; it was the result of a deliberate decision by Congress to subordinate such concerns to the achievement of health goals. Exasperated by the lack of significant progress toward dealing with the problem of air pollution under the Air Quality Act of 1967, 81 Stat. 485, and prior legislation, Congress abandoned the approach of offering suggestions and setting goals in favor of “taking a stick to the States in the form of the Clean Air Amendments of 1970 * * *.” Train v. Natural Resources Defense Council, Inc., supra, 421 U.S. at 64, 95 S.Ct. at 1474; see Union Electric Co. v. EPA, 427 U.S. 246, 256-257, 96 S.Ct. 2518, 2525, 49 L.Ed.2d 474 (1976). Congress was well aware that, together with Sections 108 and 110, Section 109 imposes requirements of a “technology-forcing” character. Id. at 257, 96 S.Ct. at 2525; Train v. Natural Resources Defense Council, Inc., supra, 421 U.S. at 91, 95 S.Ct. at 1487; Ethyl Corp. v. EPA, supra, 541 F.2d at 14. The Senate Report on the 1970 Amendments declared:

The protection of public health — as required by the national ambient air quality standards * * * — will require major action throughout the Nation. Many facilities will require major investments in new technology and new processes. Some facilities will need altered operating procedures * * . Some may be closed.

* * * * * *

In the Committee discussions, considerable concern was expressed regarding the use of the concept of technical feasibility as the basis of ambient air standards. The Committee determined that 1) the health of people is more important than the question of whether the early achievement of ambient air quality standards protective of health is technically feasible; and, 2) the growth of pollution load in many areas, even with application of available technology, would still be deleterious to public health.

The Report concluded:

Therefore, the Committee determined that existing sources of pollutants either should meet the standard of the law or be closed down, and in addition that new sources should be controlled to the maximum extent possible to prevent atmospheric emissions.

S.Rep.No.91-1196, supra, at 2-3. It is difficult to reconcile these statements of legislative intent with St. Joe’s claim that Congress wanted the Administrator to consider economic and technological feasibility in setting air quality standards. The “technology-forcing” requirements of the Act “are expressly designed to force regulated sources to develop pollution control devices that might at the time appear to be economically or technologically infeasible.” Union Electric Co. v. EPA, supra, 427 U.S. at 257, 96 S.Ct. at 2525.

*1150Furthermore, St. Joe’s attempt to find a mandate for the Administrator to consider economic or technological feasibility in the Act’s “adequate margin of safety” requirement is to no avail. The Senate Report explained the purpose of the margin of safety requirement:

Margins of safety are essential to any health-related environmental standards if a reasonable degree of protection is to be provided against hazards which research has not yet identified.

S.Rep.No.91-1196, supra, at 10. We are unable to discern here any congressional intent to require, or even permit, the Administrator to consider economic or technological factors in promulgating air quality standards. And when Congress directs an agency to consider only certain factors in reaching an administrative decision, the agency is not free to trespass beyond the bounds of its statutory authority by taking other factors into account. American Overseas Airlines, Inc. v. CAB, 254 F.2d 744, 748 (D.C.Cir.1958). A policy choice such as this is one which only Congress, not the courts and not EPA, can make. Indeed, the debates on the Act indicate that Congress was quite conscious of this fact. For example, Senator Muskie, one of the prime architects of the Act, in speaking about the automobile emission standards and the automobile industry, noted:

* * * I think that we have an obligation to lay down the standards and requirements of this bill.

I think that the industry has an obligation to try to meet them. If, in due course, it cannot, then it should come to Congress and share with the Congress— the representatives of the people — the need to modify the policy.

1 Legislative History of Clean Air Act Amendments of 1970 at 232 (Senate Debate on S. 4358, Sept. 21,1970) (hereinafter Leg-is.Hist.). See also id. at 236-240.38 In the same manner, if there is a problem with the economic or technological feasibility of the lead standards, St. Joe, or any other party affected by the standards, must take its case to Congress, the only institution with the authority to remedy the problem.39

It may well be that underlying. St. Joe’s argument is its feeling that Congress could not or should not have intended this result, and that this court should supply relief by grafting a requirement of economic or technological feasibility onto the statute. The Supreme Court confronted a similar suggestion in the Tellico Dam case. TVA v. Hill, 437 U.S. 153, 98 S.Ct. 2279, 57 L.Ed.2d 117 (1978). There TVA argued that the Endangered Species Act should not be construed to prevent operation of the dam since it had *1151already been completed at a cost of approximately $100 million, Congress had appropriated funds for the dam even after the Act was passed, and the species at risk — the snail darter — was relatively unimportant and ways might ultimately be found to save it. The Court rejected the invitation to “view the * * * Act ‘reasonably,’ and hence shape a remedy that ‘accords with some modicum of common sense and the public weal.’ ” Id. at 194, 98 S.Ct. at 2301. The Court pointed out:

Our individual appraisal of the wisdom or unwisdom of a particular course consciously selected by the Congress is to be put aside in the process of interpreting a statute. Once the meaning of an enactment is discerned and its constitutionality determined, the judicial process comes to an end. We do not'sit as a committee of review, nor are we vested with the power of veto. * * *

Id. at 194-195, 98 S.Ct. at 2302. Without suggesting that the Administrator’s interpretation of the statute does not “accord[] with some modicum of common sense and the public weal,” the Supreme Court’s admonition reinforces our decision to reject St. Joe’s invitation to construe the statute as requiring the Administrator to consider economic and technological feasibility in promulgating air quality standards.

For its part, LIA maintains that its claim that the Administrator exceeded the bounds of his statutory authority does not depend on the supposition that he is required, or even permitted, to consider economic and technological feasibility in setting air quality standards. LIA contends that, instead, its argument is based on the fact that Congress itself was concerned about the question of the economic feasibility of compliance with air quality standards, a concern which was reflected in the statute it enacted. According to LIA, Congress was mindful of the possibility that air quality standards which are too stringent could cause severe economic dislocation. For this reason it only granted the Administrator authority to adopt air quality standards which are “designed to protect the public from adverse health effects that are clearly harmfulf.]”40 LIA finds support for its interpretation of congressional intent in various portions of the legislative history of the Act. For example, it notes that the Senate Report on the 1970 legislation states that EPA “would be required to set a national minimum standard of air quality,” S.Rep. No.91-1196, supra, at 10 (emphasis added), and that Senator Muskie pointed out during the floor debates that “air quality standards which will protect the public must be set as minimum standards * * 1 Legis. Hist., supra, at 125 (emphasis added).41 LIA then argues that the Administrator based the lead air quality standards on protecting children from “subclinical” effects *1152of lead exposure which have not been shown to be harmful to health, that in so doing the Administrator ignored the clear limitation that Congress imposed on his standard-setting powers, and that the Administrator’s action will in fact cause the very result that Congress was so concerned about avoiding.

LIA’s argument appears to touch on two issues. The first concerns the type of health effects on which the Administrator may base air quality standards, i.e., the point at which the Administrator’s regulatory authority may be exercised. This issue, as LIA suggests, does concern the limits that the Act, and its legislative history, may place on the Administrator’s authority. The second issue appears to be more in the nature of an evidentiary question: whether or not the evidence in the record substantiates the Administrator’s claim that the health effects on which the standards were based do in fact satisfy the requirements of the Act. Although these two issues are closely related, they are conceptually distinct, and they are best examined separately-

Section 109(b) does not specify precisely what Congress had in mind when it directed the Administrator to prescribe air quality standards that are “requisite to protect the public health.” The legislative history of the Act does, however, provide some guidance. The Senate Report explains that the goal of the air quality standards must be to ensure that the public is protected from “adverse health effects.” S.Rep.No.91-1196, supra, at 10. And the report is particularly careful to note that especially sensitive persons such as asthmatics and emphy-sematics are included within the group that must be protected. It is on the interpretation of the phrase “adverse health effects” that the disagreement between LIA and EPA about the limits of the Administrator’s statutory authority appears to be based. LIA argues that the legislative history of the Act indicates that Congress only intended to protect the public against effects which are known to be clearly harmful to health,42 maintaining that this limitation on the Administrator’s statutory authority is necessary to ensure that the standards are not set at a level which is more stringent than Congress contemplated. The Administrator, on the other hand, agrees that primary air quality standards must be based on protecting the public from “adverse health effects,” but argues that the meaning LIA assigns to that phrase is too limited. In particular, the Administrator contends that LIA’s interpretation is inconsistent with the precautionary nature of the statute, and will frustrate Congress’ intent in requiring promulgation of air quality standards.

The Administrator begins by pointing out that the Act’s stated goal is “to protect and enhance the quality of the Nation’s air resources so as to promote the public health and welfare and the productive capacity of its population[.]” Section 101(b)(1), 42 U.S.C. § 7401(b)(1). This goal was reaffirmed in the 1977 Amendments. For example, the House Report accompanying the Amendments states that one of its purposes is “[t]o emphasize the preventive or precautionary nature of the act, i.e., to assure that regulatory action can effectively prevent harm before it occurs; to emphasize the predominant value of protection of public health[.]” H.R.Rep.No.95-294, 95th Cong., 1st Sess. 49 (1977). The Administrator notes that protecting the public from harmful effects requires decisions about exactly what these harms are, a task Congress left to his judgment. He notes that the task of making these decisions is complicated by the absence of any clear thresholds above which there are adverse effects and below which there are none. Rather, as scientific knowledge expands and analytical techniques are improved, new information is uncovered which indicates that pollution levels that were once considered harmless are not in fact harmless. Congress, the Administrator argues, was conscious of this *1153problem, and left these decisions to his judgment partly for this reason.43 In such situations the perspective that is brought to bear on the problem plays a crucial role in determining what decisions are made. Because it realizéd this, Congress, the Administrator maintains, directed him to err on the side of caution in making these judgments. First, Congress made it abundantly clear that considerations of economic or technological feasibility are to be subordinated to the goal of protecting the public health by prohibiting any consideration of such factors. Second, it specified that the air quality standards must also protect individuals who are particularly sensitive to the effects of pollution. Third, it required that the standards be set at a level at which there is “an absence of adverse effect” on these sensitive individuals. Finally, it specifically directed the Administrator to allow an adequate margin of safety in setting primary air quality standards in order to provide some protection against effects that research has not yet uncovered. The Administrator contends that these indicia of congressional intent, the precautionary nature of the statutory mandate to protect the public health, the broad discretion Congress gave him to decide what effects to protect against, and the uncertainty that must be part of any attempt to determine the health effects of air pollution, are all extremely difficult to reconcile with LIA’s suggestion that he can only set standards which are designed to protect against effects which are known to be clearly harmful to health.

We agree that LIA’s interpretation of the statute is at odds with Congress’ directives to the Administrator. As a preliminary matter, though it denies this, LIA does at times seem to be arguing, along with St. Joe, that the Administrator should have considered economic and technological feasibility in setting the standards,44 a claim that must be rejected for reasons we have already stated.45 Be that as it may, it is not immediately clear why LIA expects this court to impose limits on the Administrator’s authority which, so far as we can tell, Congress did not. The Senate Report explains that the Administrator is to set standards which ensure that there is “an absence of adverse effects.” The Administrator maintains that the lead standards are designed to do just that, a claim we will examine in due course. But LIA would *1154require a further showing — that the effects on which the standards were based are. clearly harmful or clearly adverse. We cannot, however, find the source of this further restriction that LIA would impose on the Administrator’s authority.46 It may be that it reflects LIA’s view that the Administrator must show that there is a “medical consensus that [the effects on which the standards were based] are harmful * 47 If so, LIA is seriously mistaken. This court has previously noted that some uncertainty about the health effects of air pollution is inevitable.48 And we pointed out that “[a]waiting certainty will often allow for only reactive, not preventive regulat[ory action].” Ethyl Corp. v. EPA, supra, 541 F.2d at 25. Congress apparently shares this view; it specifically directed the Administrator to allow an adequate margin of safety to protect against effects which have not yet been uncovered by research and effects whose medical significance is a matter of disagreement.49 This court has previously acknowledged the role of the margin of safety requirement. In Environmental Defense Fund v. EPA, 598 F.2d 62, 81 (D.C. Cir.1978), we pointed out that “[i]f administrative responsibility to protect against unknown dangers presents a difficult task, indeed, a veritable paradox — calling as it does for knowledge of that which is unknown — then, the term ‘margin of safety’ is Congress’s directive that means be found to carry out the task and to reconcile the paradox.” Moreover, it is significant that Congress has recently acknowledged that more often than not the “margins of safety” that are incorporated into air quality standards turn out to be very modest or nonexistent, as new information reveals adverse health effects at pollution levels once thought to be harmless. See H.R.Rep.No. 95-294, supra, at 103-117. Congress’ directive to the Administrator to allow an “adequate margin of safety” alone plainly *1155refutes any suggestion that the Administrator is only authorized to set primary air quality standards which are designed to protect against health effects that are known to be clearly harmful.

Furthermore, we agree with the Administrator that requiring EPA to wait until it can conclusively demonstrate that a particular effect is adverse to health before it acts is inconsistent with both the Act’s precautionary and preventive orientation and the nature of the Administrator’s statutory responsibilities. Congress provided that the Administrator is to use his judgment in setting air quality standards precisely to permit him to act in the face of uncertainty.50 And as we read the statutory provisions and the legislative history, Congress directed the Administrator to err on the side of caution in making the necessary decisions. We see no reason why this court should put a gloss on Congress’ scheme by requiring the Administrator to show that there is a medical consensus that the effects on which the lead standards were based are “clearly harmful to health.” All that is required by the statutory scheme is evidence in the record which substantiates his conclusions about the health effects on which the standards were based.51 Accord*1156ingly, we reject LIA’s claim that the Administrator exceeded his statutory authority and turn to LIA’s challenge to the evi-dentiary basis for the Administrator’s decisions.

VI. HEALTH BASIS FOR THE LEAD STANDARDS

LIA does not question a number of the steps in the Administrator’s analysis. It does not disagree with his selection of children between the ages of one and five years as the target population, or the decision to set a standard that would keep 99.5 percent of the children below the maximum safe individual blood lead level. In addition, LIA does not challenge the Administrator’s suggestion that the standards should be based on an assumption that non-air sources contribute 12 ug Pb/dl to blood lead levels. LIA does, however, challenge other key elements in the Administrator’s analysis.

A. Maximum Safe Individual Blood Lead Level

LIA attacks the Administrator’s determination that 30 ug Pb/dl should be considered the maximum safe individual blood lead level for children, maintaining that there is no evidence in the record indicating that children suffer any health effects that can be considered adverse at this blood lead level. As previously noted,52 the Administrator’s selection was based on his finding that EP elevation at 30 ug Pb/dl is the first adverse health effect of lead exposure, and his determination that a maximum safe individual blood lead level of 30 ug Pb/dl will allow an adequate margin of safety in protecting children against the more serious effects of lead exposure — anemia, symptoms of which appear at blood lead levels of 40 ug Pb/dl and central nervous system deficits which begin to occur at blood lead levels of 50 ug Pb/dl.

LIA challenges each of these findings. First, it contends that nothing in the record supports the suggestion that EP elevation at 30 ug Pb/dl is harmful to health, arguing that EP elevation is a mere “subclinical effect” — a biological response to lead exposure — which is without health significance, and noting that a number of its experts brought this matter to EPA’s attention in their comments on the proposed standards.53 In LIA’s view, the Administrator did not explain precisely how impairment of heme synthesis at blood lead levels of 30 ug Pb/dl adversely affects the health of children.54 Second, LIA challenges the Administrator’s determination that a maximum safe individual blood lead level of 30 ug Pb/dl is justified by the need to allow an adequate margin of safety in protecting children against anemia and central nervous system deficits. It maintains that the evidence in the record does not support the Administrator’s conclusion that the blood lead threshold for the symptoms of anemia in children is 40 ug Pb/dl. LIA claims that this error was brought to the Administrator’s attention by comments on the proposed standard, but that he failed to respond to these comments, thereby violating the statutory provision requiring him to respond to “significant comments, criticisms, and new data submitted * * * during the comment period.”55 Third, LIA contends that the *1157preamble to the final regulations does not state the basis for the Administrator’s finding that central nervous system deficits occur in children at blood lead levels of 50 ug Pb/dl, thereby precluding this court from being able to test the soundness of this determination. Finally, LIA argues that even if it were to concede that EPA’s conclusions about the blood lead thresholds for anemia and central nervous system deficits are correct, there is still no explanation of why the Administrator concluded that a maximum individual safe blood level of 30 ug Pb/dl — rather than 35 ug Pb/dl, for example — is necessary to provide an adequate margin of safety against these effects.

Our review of the record persuades us that there is adequate support for each of the Administrator’s conclusions about the health effects of lead exposure and, consequently, that LIA’s challenges to the evi-dentiary support for these findings must be rejected. Under the statutory scheme enacted by Congress, the Criteria Document prepared with respect to each pollutant is to provide the scientific basis for promulgation of air quality standards for the pollutant. We have already noted that the Lead Criteria Document was the product of a process that allowed the rigorous scientific and public review that are essential to the preparation of a document “accurately reflecting] the latest scientific knowledge useful in indicating the kind and extent of all identifiable effects [of lead exposure] on [the] public health * * 56 In our view, the Criteria Document provides ample support for the Administrator’s findings.

The Criteria Document concluded that EP elevation, which begins in children at blood levels of 15-20 ug Pb/dl, is one of the more significant effects of low level lead exposure because it indicates that lead has already begun to affect basic biological functions in the body. We have already examined in some detail the findings that led the Criteria Document to the conclusion that EP elevation is an indication of a physiological impairment which affects human health.57 Briefly stated, it concluded that EP elevation indicates an impairment of the functioning of the mitochondria58; that EP elevation indicates that lead has begun to affect the process of heme synthesis; that EP elevation may indicate a reduction in any reserve capacity there might be in the heme synthesis system; and that lead’s impairment of heme synthesis in the blood suggests that lead may also interfere with production of heme proteins in other organ systems. Relying on the Criteria Document’s discussion, as well as other evidence in the record, the Administrator made a judgment that for purposes of setting air quality standards for lead, EP elevation at 30 ug Pb/dl must be considered the first adverse effect on the health of children, and he determined that the maximum safe individual blood lead level should be no higher than 30 ug Pb/dl.

The Administrator’s judgment echoes the consensus of a group of clinicians who, in 1975, participated in preparation of a statement issued by the Center For Disease Control and endorsed by the American Academy of Pediatrics. These experts agreed that EP elevation “should be used as an indicator of a significant and worrisome body burden of lead.” CD 11-14, JA 1236. Moreover, the Center For Disease Control uses EP elevation at 30 ug Pb/dl as the cutoff point for detection of lead poisoning in children in its screening programs,59 a *1158factor that influenced the Administrator’s decision. 43 Fed.Reg. 46253, JA 2955. The Administrator’s reliance on this was, in our view, entirely appropriate. While the Center has indicated that children with a blood lead concentration of 30 ug Pb/dl have a lead problem,60 it is also significant that the Center’s screening program has much the same preventive and precautionary approach that Congress directed the Administrator to apply to the task of setting air quality standards designed to protect the public health.61 Further support for the Administrator’s decisions is provided by the testimony of various medical and other scientific experts who participated in the lead standards rulemaking. These experts endorsed the Administrator’s (and Criteria Document’s) conclusions about the effects of low level lead exposures, and agreed with his assessment of the health significance of these effects.62 The Administrator’s decision is, of course, precisely the sort of issue that Congress specifically left to his judgment, and where there is evidence in the record which supports these judgments, this court is not at liberty to substitute its judgment for the Administrator’s. In this instance the Administrator has acted properly under the terms of the statute. He has explained his factual findings and policy judgments, and there is an adequate basis in the record for these decisions. No more is required of him.

LIA’s challenge to the Administrator’s findings concerning the health significance of EP elevation also stresses that this phenomenon is only a “subclinical” effect. But the clinical/subclinical distinction has little to do with the question whether a particular effect is properly viewed as adverse to health. Rather, the distinction pertains to the means through which the particular effect may be detected: observation or physical examination in the case of clinical effects, and laboratory tests in the case of subclinical effects.63 Thus describing a particular effect as a “subclinical” effect in no way implies that it is improper to consider it adverse to health.64 While EP elevation may not be readily identifiable as a sign of disease, the Administrator properly concluded that it indicates a lead-related interference with basic biological functions. Expert medical testimony in the record confirms that the modern trend in preventive medicine is to detect health problems in their “subclinical” stages, and thereupon to take corrective action.65 Moreover, as we have already noted, the Center For Disease Control uses the same “subclinical” effect as the key indicator of the need for medical intervention in its lead poisoning screening program. The accepted use of this “sub-*1159clinical” effect to determine the need for medical observation or intervention properly influenced the Administrator’s decision.66 Thus the fact that the effects the Administrator relied on in setting the lead standards are “subclinical” does not detract from their significance for human health, or make them an improper basis for setting air quality standards.

We also reject LIA’s claim that the evidence in the record does not support the Administrator’s determination that the blood lead threshold for symptoms of anemia in children is 40 ug Pb/dl. According to the Criteria Document, the onset of anemia is marked by a decline in the level of hemoglobin per unit of blood, CD 13-4, JA 1334. And the Criteria Document states that “[rjeports on children indicate that statistically significant decreases in hemoglobin levels begin to appear at a blood lead level of 40 ug [Pb]/dl or somewhat below.” Id. At another point the Criteria Document explains, “In children, a threshold blood lead level for production of * * * symptoms of anemia is approximately 40 ug Pb/dl * * CD 1-6, JA 1110. See id. 11-13, JA 1235; JA 2583-2585 (Needleman and Piomelli). The Criteria Document’s conclusions were reached after a review of various studies that have examined the subject, and we cannot, in light of these findings, say that the Administrator’s decision about the threshold blood lead level for the symptoms of anemia in children does not have an adequate basis in the record.67

Finally, our examination of the record also reveals ample support for the Administrator’s determination that lead-induced central nervous system deficits begin to occur in children at blood lead levels of 50 ug Pb/dl. The central nervous system damage about which the Administrator was concerned was not the severe brain damage that can occur at relatively high levels of lead exposure — 80-100 ug Pb/dl.68 Rather, his focus was on more subtle and largely irreversible neurological and behavioral impairment that has been detected in children at lower blood lead levels, 43 Fed.Reg. 46253, JA 2955.69 The Criteria Document candidly admitted that “[t]he literature on this subject is somewhat limited and controversial,” but it was nevertheless able to conclude that “certain statements [can] be made about the possible hazard of low to moderate lead exposure levels.” CD 11-18, JA 1240. The conclusion it reached, after a detailed review of various studies that have examined the subject, was that:

[The] evidence tends to confirm that some type of neural damage does exist in asymptomatic children, and not necessarily only at very high level of blood lead. The body of studies on low- or moderate-level lead effects on neurobehavioral functions * * * present overall a rather impressive array of data pointing to that conclusion. Several well-controlled studies have found effects that are clearly statistically significant, whereas others have found nonsignificant but borderline effects. Even some studies reporting generally nonsignificant *1160findings at times contain data confirming statistically significant effects, which the authors attribute to various extraneous factors. * * *

CD 11-26, JA 1248. The Criteria Document reported that the blood lead levels associated with these neurobehavioral deficits are 50-60 ug Pb/dl. Id. These conclusions were endorsed by several of the experts who participated in the lead standards rule-making proceedings, including one of LIA’s experts.70 Some of these experts even suggested that these effects may occur at blood lead levels lower than the levels indicated by the Criteria Document.71 Contrary to LIA’s suggestion, the evidence in the Criteria Document and the testimony of the experts provides an adequate basis for this court to undertake a review of the Administrator’s findings concerning these effects. Accordingly, we reject LIA’s challenge to the Administrator’s conclusion that central nervous system deficits begin to occur in children at blood lead levels of 50 ug Pb/dl.

Our conclusion that there is ample support for the Administrator’s determination that EP elevation at 30 ug Pb/dl is the first adverse health effect that children experience as a result of lead exposure is, of course, sufficient to sustain his selection of 30 ug Pb/dl as the maximum safe individual blood lead level. Given this, we cannot say that his further determination that a maximum safe individual blood lead level of 30 ug Pb/dl would in addition provide protection against the more serious adverse health effects of lead exposure was irrational.

To be sure, the Administrator’s conclusions were not unchallenged; both LIA and the Administrator are able to point to an impressive array of experts supporting each of their respective positions.72 However, disagreement among the experts is inevitable when the issues involved are at the “very frontiers of scientific knowledge,” and such disagreement does not preclude us from finding that the Administrator’s decisions are adequately supported by the evidence in the record. It may be that LIA expects this court to conclude that LIA’s experts are right, and the experts whose testimony supports the Administrator are wrong.73 If so, LIA has seriously misconceived our role as a reviewing court. It is not our function to resolve disagreement among the experts or to judge the merits of competing expert views. AFL-CIO v. Marshall, 617 F.2d 636, 651 & n.66 (D.C.Cir. 1979); cf. Hercules Inc. v. EPA, 598 F.2d 91, 115 (D.C.Cir.1978) (“[c]hoice among scientific test data is precisely the type of judgment that must be made by EPA, not this court”). Our task is the limited one of ascertaining that the choices made by the Administrator were reasonable and supported by the record. Ethyl Corp. v. EPA, supra, 541 F.2d at 35-36. That the evidence in the record may also support other conclusions, even those that are inconsistent with the Administrator’s,74 does not prevent us from concluding that his decisions were rational and supported by the record. AFL-CIO v. Marshall, supra, 617 F.2d at 651 n.66; Environmental Defense Fund, Inc. v. EPA, 510 F.2d 1292, 1298 (D.C.Cir. 1975); accord, Bayside Enterprises, Inc. v. NLRB, 425 U.S. 298, 302, 97 S.Ct. 576, 579, 50 L.Ed.2d 494 (1976). Cf. Universal Camera Corp. v. NLRB, 340 U.S. 474, 488, 71 S.Ct. 456, 465, 95 L.Ed. 456 (1951) (“a court may [not] displace the [agency’s] choice between two fairly conflicting views, even though the court would justifiably have *1161made a different choice had the matter been before it de novo ”).

Having determined that we must uphold the Administrator’s decisions concerning the health effects that are the basis for the lead standards, we turn to petitioners’ other challenges to the Administrator’s analysis.

B. Margin of Safety

Both LIA and St. Joe argue that the Administrator erred by including multiple allowances for margins of safety in his calculation of the lead standards. Petitioners note that the statute directs the Administrator to allow an “adequate margin of safety” in setting primary air quality standards, and they maintain that as a matter of statutory construction the Administrator may not interpret “margin” of safety to mean “margins” of safety. In petitioners’ view, the Administrator in fact did just this insofar as he made allowances for margins of safety at several points in his analysis. They argue that margin of safety allowances were reflected in the choice of the maximum safe individual blood lead level for children, in the decision to place 99.5 percent of the target population group below that blood lead level, in the selection of an air lead/blood lead ratio at 1:2, and in the Administrator’s estimate of the contribution to blood lead levels that should be attributed to non-air sources.75 The net result of these multiple allowances for margins of safety, petitioners contend, was a standard far more stringent than is necessary to protect the public health. St. Joe suggests that EPA should have adopted an approach which required decisions on:

1) The maximum level of lead in air which is protective of health; i.e., a threshold beyond which the public health is not protected; and

2) An adequate margin of safety by which the level which is protective of health must be reduced.[76]

EPA responds by maintaining that allowances for a margin of safety were made only at two points in its analysis: in the selection of a maximum safe individual blood lead level of 30 ug Pb/dl and in the decision to set a standard designed to keep 99.5 percent of the target population below that blood lead level. It argues that the statutory requirement of a margin of safety does not mandate adoption of the method suggested by St. Joe. Rather, EPA suggests, it indicates the precautionary orientation the Administrator is to bring to bear on the task of setting air quality standards. How conservative he must be in making particular judgments must, the Agency maintains, depend on such factors as the amount of uncertainty involved, the size of the population affected, and the severity of the effect. EPA argues that petitioners’ claims about multiple allowances for margins of safety indicate that they have failed to recognize the difference between providing for a margin of safety and making a scientific judgment in the face of conflicting evidence.77

We agree with the Administrator that nothing in the statutory scheme or the legislative history requires him to adopt the margin of safety approach suggested by St. Joe.78 Adding the margin of safety at the end of the analysis is one approach, but it is not the only possible method. Indeed, the *1162Administrator considered this approach but decided against it because of complications raised by the multiple sources of lead exposure.79 The choice between these possible approaches is a policy choice of the type that Congress specifically left to the Administrator’s judgment. This court must allow him the discretion to determine which approach will best fulfill the goals of the Act. As we pointed our in Hercules Inc. v. EPA, supra, 598 F.2d at 108, “Decision between the alternatives is a quintessential policy judgment within the discretion of EPA. We cannot accept [the] notion that the administrator of the agency created to protect the environment lack[s] even the capability to exercise the discretion with which he was entrusted by Congress.” (Emphasis in original.)80 Where, as here, the Administrator has provided an explanation of why he chose one method rather than another, and this explanation and his choice are not irrational, we must accept his decision. See Industrial Union Dep’t, AFL-CIO v. Hodgson, supra, 499 F.2d at 475-476.

We also agree with the Administrator’s suggestion that petitioners have ignored the distinction between scientific judgments based on the available evidence and allowances for margins of safety. In every instance in which the Administrator’s judgment on a particular issue differed from petitioners’ they attributed his decision to an allowance for a margin of safety. To be sure, there is no bright line that divides these two types of decisions, but they are nonetheless conceptually distinct. In any event, whatever the nature of the decision, the real test, as petitioners recognize,81 is whether the decision is reasonable when examined in light of the evidence in the record. We have already found that at least one of the decisions that the petitioners attribute to an allowance for a margin of safety — the selection of the maximum safe individual blood lead level for children — satisfies this test. Accordingly, we turn to petitioners’ claims that the other steps in the Administrator’s analysis cannot withstand critical scrutiny.

VII. ALLEGED ARBITRARY AND CAPRICIOUS DECISIONS

Petitioners contend that a number of the findings which constitute the very core of the Administrator’s analysis violate one or more of the decisionmaking requirements of the Act. While arguing that each of these violations is sufficient ground for remand of the regulations to EPA, LIA maintains that “cumulatively they paint a picture of an agency that had prejudged the result from the very outset and was bent on adhering to its original proposal no matter what the evidence showed, the very converse of the fair and impartial rulemaking to which litigants * * * are entitled.”82

A. Air Lead/Blood Lead Ratio

LIA contends that the Administrator’s choice of an air lead/blood lead ratio of 1:2 as the appropriate ratio for calculating the lead standards was arbitrary and capricious. LIA’s claim is largely based on its disagreement with the Administrator’s interpretation of the results of three studies that have examined the relationship between air lead exposure and blood lead levels. It argues that the Administrator’s analysis of these studies is inconsistent and designed solely to support his decision to arrive at an air quality standard of 1.5 ug Pb/m8.83 In addition, *1163LIA contends that the Administrator erred by calculating the air lead/blood lead ratio on the basis of studies involving both adults and children, when the standards are designed to protect children. LIA maintains that the Administrator would have arrived at a ratio of 1:1.3, had he focused solely on the studies involving children.

We do not agree that the Administrator’s selection of an air lead/blood lead ratio of 1:2 was arbitrary or capricious. The Criteria Document reported that air lead/blood lead ratios for the whole population, adults as well as children, range between 1:1 and 1:2, with children at the upper end of the range or even slightly above it. CD 12-38, JA 1325. And the range of ratios for children reported by the studies that were reviewed in the Criteria Document was 1:1.2 to 1:2.3, CD 12-25 (Table 12-28), JA 1312 (Table 12-28).84 The Administrator’s choice of a ratio of 1:2 for purposes of calculating the lead standards is consistent with each of these findings. Moreover, the Administrator calculated that each of three particularly relevant and well-documented studies that were reviewed by the Criteria Document suggested an air lead/blood lead ratio close to 1:2. 43 Fed.Reg. 46250, 46254, JA 2954, 2956.85 Finally, the Administrator’s choice of a ratio of 1:2 was endorsed by several experts who participated in the rulemaking proceedings.86 Indeed, the issue of the proper relationship between air lead exposure and blood lead levels was extensively discussed in the comments on the initial drafts of the Criteria Document, with several experts severely criticizing the suggestions in early drafts that the appropriate ratio is 1:1.87 Given all the evidence in the record which supports the Administrator’s choice of a ratio of 1:2, we would be exceeding the scope of our reviewing function if we were to agree with LIA’s suggestion that the Administrator’s decision was either arbitrary or capricious.88

*1164B. Changes in Method

LIA next argues that the Administrator contravened the decisionmaking requirements of Section 307(d), 42 U.S.C. § 7607(d), by failing to explain the reasons for a change in the method he used in calculating the lead standards between the proposed and the final standards.89 LIA correctly points out that the final standard was based on an adverse health effects threshold of 30 ug Pb/dl, whereas the proposed standards had been based on a threshold of 15 ug Pb/dl. It notes that one reason why both the proposed and the final standards nevertheless arrived at an air quality standard of 1.5 ug Pb/m3 was that the Administrator employed different statistical procedures in determining the target mean population blood level for the two standards.90 While intimating that the change in methods was not unrelated to EPA’s desire to arrive at a final standard of 1.5 ug Pb/m3, LIA contends that the Administrator did not explain the reasons for this change in method as required by the Act. LIA further argues that the Administrator failed to reconcile his adoption of the statistical procedure used in calculating the final standard with his earlier suggestion (in the proposed standards) that this method “may overestimate the degree to which the population mean should be below the threshold blood lead level.” 42 Fed.Reg. 63079, JA 1483. LIA maintains that the Administrator should either have corrected for the use of such an overprotective procedure or explained the reasons why he chose not to do so.

We find LIA’s contentions to be without substantial merit. In evaluating the significance of these claims, we cannot help noticing that in spite of the misgivings the Administrator had expressed about the lognormal statistical procedure, both LIA and its experts endorsed the use of this procedure in their comments on the proposed standards, and in fact used it to calculate the alternative standards that they recommended.91 LIA’s newly discovered objection to the use of this procedure thus really seems directed at the result it produced, rather than the mere fact that the Administrator used it.92 Be that as it may, we are satisfied that the Administrator complied with the requirements of Section 307(d). At the time he issued the proposed standards the Administrator informed the public that use of lognormal statistical procedures was an alternative approach to the method he had employed in calculating the proposed standards, and he candidly explained that he had some reservations about the procedure. 42 Fed.Reg. 63079, JA 1483. A fair reading of the Administrator’s discussion of the issue in the final regulations *1165suggests that the comments on the proposed standards, including the comments submitted by LIA and its experts, persuaded him to reexamine his analysis, and to conclude that his earlier misgivings about the lognormal procedure were exaggerated. 43 Fed.Reg. 46252-46253, JA 2954-2955. And we are satisfied that it is possible to discern the reasons why the Administrator decided to adopt this procedure from his discussion. Id.93 Accordingly, we must conclude that his discussion of the alternative methods and the reasons for the change in his approach were more than adequate to comply with the requirements of Section 307(d).

Finally, we have uncovered nothing in the record that indicates that the procedure is unreliable, or that the Administrator’s decision to use it was unreasonable. Moreover, so far as we can tell, at no time during the course of the rulemaking proceedings did LIA raise any objections to, or even express any reservations about, the lognormal statistical procedure, this in spite of the misgivings the Administrator expressed in the proposed standards.94 LIA did not even mention this issue in the petition it filed with EPA for reconsideration and stay of the lead standards. In these circumstances, remanding the regulations to EPA is totally unwarranted. LIA would do well to remember the Supreme Court’s admonition that “administrative proceedings should not be a game or a forum to engage in unjustified obstructionism by making [no reference to an issue] and then, after failing to * * * bring the matter to the agency’s attention, seeking to have that agency determination vacated on the ground that the agency failed to consider [the matter] * * *." Vermont Yankee Nuclear Power Corp. v. Natural Resources Defense Council, Inc., 435 U.S. 519, 553-554, 98 S.Ct. 1197, 1217, 55 L.Ed.2d 460 (1978).

C. Insoluble and Non-Respirable Particles

Both LIA and St. Joe argue that the Administrator acted arbitrarily and capriciously by refusing to exclude lead particles that are insoluble, or non-respirable because of their size, from the ambient air quality standards, despite the fact that this issue was brought to his attention. LIA points out that the Criteria Document suggests that particle size and solubility may affect absorption of lead emissions through the lungs.95 Petitioners conclude from this that the Administrator should have excluded insoluble and non-respirable particles from the standards, arguing that his failure to do so is particularly unfair to stationary emission sources because a significant portion of lead emissions from such sources are of this nature. St. Joe maintains that exclusion of these particles from the standards would minimize the adverse economic impact that the lead standards will have on the industry. And LIA argues that the Administrator’s response to its comments on this issue *1166ignored the question of solubility, thereby violating the requirement that he respond to significant comments and criticisms.96

The Administrator offers a number of justifications for his decision not to exclude insoluble particles from the lead standards. First, the Administration dismisses LIA’s suggestion that he was required to respond to the claim that certain lead particles are insoluble, pointing out that very little evidence was presented to support this claim and that such evidence as there was focused on high level occupational exposures and was therefore of little relevance to the task of setting air quality standards for the whole population.97 He further notes that studies discussed in the Criteria Document indicate that non-industry-employed populations living in the vicinity of smelters show high blood lead levels and severe health impairment98 and he argues that these revelations refute the suggestion that a significant portion of emissions from such sources is insoluble and cannot be absorbed into the blood. Finally, EPA contends that even if the Administrator erred in not explaining why he rejected the suggestion that he exclude insoluble particles from the standard, this was at most a harmless procedural error which is not ground for remand of the lead standards.

The Administrator also offers several justifications for his decision not to exclude from the lead standards particles which are supposedly non-respirable because they are too large. He begins by noting that these larger particles — those that are over one micron in size — constitute only a very small percentage of overall airborne lead,99 and that the data available on particle size retention in the lungs and subsequent absorption into the bloodstream is very limited.100 Next, he points out that studies discussed in the Criteria Document and other evidence in the record indicate that these larger particles are also retained in the lungs and subsequently absorbed into the blood, although to a lesser degree than are smaller particles.101 As such, some percentage of these larger particles are in fact respirable. In addition, the Administrator notes that the Criteria Document indicates that some portion of these larger particles are cleared from the throat and lungs, swallowed, and subsequently absorbed ' into the blood through the intestines.102

These facts alone, the Administrator argues, are sufficient to sustain the decision not to distinguish between respirable and non-respirable particles. But he also points out that one other consideration played a role in the decision and provides further support for it. The Administrator explains that in areas with high concentrations of airborne lead, such as near lead smelters or major highways, much of the lead settles on the ground and may eventually become a source of human lead exposure through ingestion of lead-contaminated food or, particularly in children, placing hands and other contaminated objects in the mouth. 43 Fed.Reg. 46251, JA 2953. While acknowledging that some allowance was made for the contribution of non-air sources to blood lead levels in calculating the lead standards, the Administrator stresses that the 12 ug Pb/dl estimate is merely a minimum national average which does not reflect the true non-air contribution to blood lead levels near major emission sources.103 The Ad*1167ministrator argues that for this reason, as well as the others previously stated, the decision not to exclude non-respirable particles was based on firm evidence that these particles do in fact contribute to blood lead levels.

We find that the Administrator’s decision not to exclude insoluble particles from the lead standards was neither arbitrary nor capricious. The only information in the record on the issue of insolubility that petitioners are able to point to is a two-page discussion in St. Joe’s comments on the proposed standards,104 and even this discussion does not suggest that lead emissions from smelters or other industrial sources are insoluble. The Criteria Document merely states that the exposure of lead miners may “depend to some extent on the solubility of the lead from the ores.” CD 7-13, JA 1183.105 Given the paucity of the information presented on this question, there was nothing arbitrary or capricious about the Administrator’s decision to include such particles in the lead standards. Furthermore, LIA’s contention that the Administrator was required to respond to its suggestion that insoluble particles should be excluded from the standards borders on the ludicrous. Section 307(d)(6)(B) only requires the Administrator to respond to “significant comments.” LIA’s unsupported claim simply did not rise to the level of a comment which required a response from the Administrator.

We also conclude that the Administrator’s decision not to exclude nonrespirable particles from the lead standards was reasonable and supported by the record. The Criteria Document and other evidence in the record provide an adequate basis for his determination that some portion of these non-respirable particles are eventually absorbed into the bloodstream, and we agree with the Administrator that this fact alone is sufficient to demonstrate that his decision was neither arbitrary nor capricious.106 As we have previously said, our task is at an end once we have ascertained that the agency “has given reasoned consideration to all the material facts and issues,” Greater Boston Television Corp. v. FGC, supra, 444 F.2d at 851, and that its decision is supported by the record. In this case we are satisfied that the Administrator’s decision meets both tests.107

*1168VIII. PROCEDURAL OBJECTIONS

LIA also raises a variety of procedural objections to the lead standards rulemaking which, in its view, mandate remand of the lead standards to EPA.

A. The Needleman Study

LIA argues that EPA erred by relying on a study submitted after the close of the public comment period without first allowing interested parties an opportunity to comment on it. The “Needleman Study”108 examined the relationship between lead exposure measured by lead concentration in teeth and the psychological performance of young children. The study was first mentioned by Dr. Needleman when he referred briefly to a study he had conducted but had not yet published which, he said, indicated “that children who have a mean blood lead level in the past of 35 micrograms per deciliter are significantly impaired on a large number of psychological outcomes when compared to children who had a mean level in the past of 24 micrograms, a 10 micro-gram difference * * *.” JA 1626. The study itself was not submitted to EPA until June 7, 1978, after the close of the public comment period, and it was placed in the public docket on August 17, 1978.

LIA acknowledges that the Administrator did not mention the study in the preamble to the lead standards, but it nevertheless argues that EPA clearly relied on the study in formulating the final standards. In support of this allegation LIA notes that the preamble refers to “the possibility that nervous system damage may occur in children even without overt symptoms of lead poisoning” and the “possibility that lead exposure resulting in blood lead levels previously considered safe may in fact influence the neurological development and learning abilities of the young child.” 43 Fed.Reg. 46246,46255, JA 2948, 2957. Both these statements, LIA claims, are paraphrases of Dr. Neddleman’s findings. In addition, LIA points to a number of internal agency memoranda which purportedly show that the agency placed increasing reliance on the supposed low level neurological effects of lead exposure to justify the standard after it received the study.109

EPA disclaims any reliance on the Nee-dleman study in the formulation of the lead standards, pointing out that its conclusions about the effects of low level blood lead concentrations on the central nervous system and psychological performance were based on other evidence in the record. In addition, EPA notes that the lead standards were not in fact based on protecting children from neurological disorders at blood lead levels of 25-30 ug Pb/dl, which is what the Needleman study found, and points out that the only neurological effects that played a role in the Administrator’s analysis were the central nervous system deficits which occur at blood lead levels of 50-60 ug Pb/dl. EPA also contends that none of the intra-agency discussions of the Needleman study to which LIA refers suggests that it was being relied on in the formulation of *1169the standards.110 Finally, EPA argues that soliciting comments on the Needleman study was unnecessary because comments on the study would not have changed the standard since the standard was not based on the study.111 We agree.

In our view, LIA has not adduced any evidence to substantiate its claim that the Administrator relied on the Needleman study in formulating the lead air quality standards. We have already found that the Administrator’s conclusions about the health effects of lead exposure — including the statements in the preamble to which LIA refers — are amply supported by the evidence in the record. Accordingly, we have no reason to reject the Administrator’s disclaimer of reliance on the Needleman study.

B. Cross-Examination

LIA’s next procedural challenge stems from the denial of its request for an opportunity to cross-examine the medical and scientific witnesses who testified in support of the then-proposed standards at the public hearings on the lead standards. Acknowledging that the Act does not provide for cross-examination, LIA argues that this case presents a situation in which “constitutional constraints” or “extremely compelling circumstances,” see Vermont Yankee Nuclear Power Corp. v. Natural Resources Defense Council, Inc., supra, 435 U.S. at 543, 98 S.Ct. at 1211, justify imposition of additional procedures on an agency by a reviewing court. In support of this claim LIA first alleges that EPA failed to submit the scientific issues raised by the lead standards for review by the Independent Scientific Review Committee as required by statute,112 thereby “foreclos[ing] an independent, objective and unbiased review of the crucial medical and scientific data[.]”113 LIA further notes that this court has intimated that in some situations “cross-examination of live witnesses on a subject of critical importance which could not be adequately ventilated under the general procedures” may be appropriate even though not required by statute. International Harvester Corp. v. Ruckelshaus, 478 F.2d 615, 631 (D.C.Cir.1973). Finally, LIA contends that EPA itself demonstrated that it recognized the desirability of cross-examination on these issues by allowing “intensive and at times hostile cross-examination” of experts who testified against the proposed standards.114 LIA argues that EPA offended “fundamental notions of fairness implicit in due process,” Home Box Office, Inc. v. FCC, 567 F.2d 9, 56 (D.C.Cir.), cert. denied, 434 U.S. 829, 98 S.Ct. 111, 54 L.Ed.2d 89 (1977), by refusing to permit similar cross-examination of experts who testified in favor of the standards.

As LIA itself acknowledges, it faces an extremely heavy burden in its attempt to persuade this court to impose on EPA a procedure that is not required by statute. The Supreme Court’s decision in the Vermont Yankee case makes it absolutely clear that courts must be extremely reticent about going beyond the procedures established by Congress and requiring agencies *1170to provide additional procedures in rule-making proceedings. Judicial restraint in this matter is all the more important where, as here, Congress considered and deliberately decided against a particular procedure. Section 307(d)(5), 42 U.S.C. § 7607(d)(5), which governs the procedure at public hearings, was added by the 1977 Amendments to the Act. The House bill did in fact require an opportunity for cross-examination during public hearings, but the Senate bill did not. The Conference Committee adopted the House bill but deleted the cross-examination provision, substituting in its place a requirement that the hearing record remain open for 30 days after the close of the hearings in order to permit parties to submit rebuttal and supplemental information. See H.R.Rep.No.95-564, 95th Cong., 1st Sess. 177-178 (Conference Report) (1977). We may not ignore Congress’ judgment that the crucial issues in these standard-setting proceedings can be “adequately ventilated” without providing an opportunity for cross-examination.

But even assuming for purposes of argument that we could, in appropriate circumstances, require EPA to permit cross-examination on certain issues, LIA does not even come close to sustaining its burden of demonstrating that this is a proper case for imposition of this requirement. First, as we will show later, there simply is no truth to LIA's claim that there was no independent and objective review of the medical and other scientific issues raised by the task of setting air quality standards for lead.115 Second, we fail to see what significant additional information cross-examination would have uncovered. LIA participated in every stage of the lead standards rulemaking.116 It had ample opportunity to submit testimony and other evidence from its experts supporting its views about the health effects of lead exposure, and it did. Moreover, the statutory scheme enacted by Congress provided it with an opportunity to submit rebuttal information after the hearings on the proposed standards. In these circumstances LIA was “afforded a meaningful opportunity to be heard and to controvert the evidence. Fairness demands no more.” Ethyl Corp. v. EPA, supra, 541 F.2d at 54 n.124.

Indeed, the bankruptcy of LIA’s claim of procedural deprivation is revealed by its attempt to mischaracterize the proceedings at the public hearings on the proposed standards by claiming that its witnesses were subjected to intensive and hostile cross-examination. The procedure for the hearings was spelled out at the beginning of the hearings by the EPA official who presided over the hearings. Each participant presented a statement, and thereafter members of the panel (which included non-EPA experts) were allowed to ask clarifying questions.117 Moreover, the proceedings were very informal, the witnesses did not testify under oath, the formal rules of evidence did not apply, and most of the discussion between the participants was conducted on a first-name basis. A review of the transcript of the hearing shows that panel members did ask questions after the comments by experts who supported the proposed standards as well as after the statements by experts who testified against the standards.118 It is this exchange of ideas among peers that LIA, by a strange process of metamorphosis, transforms into “inten*1171sive and at times hostile cross-examination.” Having ourselves reviewed the transcript of the public hearings we do not agree with LIA’s characterization of what transpired.

Finally, LIA’s attempt to cloak its cross-examination claim in constitutional garb by invoking due process is to no avail. While the Supreme Court noted in Vermont Yankee that “constitutional constraints” may justify imposition of additional procedures in a rulemaking when an agency is making a “quasi-judicial” determination,119 the Court in that case rejected the suggestion that due process could require additional procedures such as an opportunity for cross-examination in a “pure” rulemaking proceeding, such as was involved in the instant case. Vermont Yankee Nuclear Power Corp. v. Natural Resources Defense Council, Inc., supra, 435 U.S. at 542 & n.16, 98 S.Ct. at 1211. Accordingly, we must abide by the Supreme Court’s admonition that we may not “stray beyond the judicial province to explore the procedural format or to impose upon the agency [our] own notion of which procedures are ‘best’ or most likely to further some vague, undefined public good.” Id. at 549, 98 S.Ct. at 1214.120 EPA’s hearing procedures complied with the Act’s requirements. Nothing more was required.

C. Independent Scientific Review Committee

One of the changes made by the 1977 Amendments to the Clean Air Act was the establishment of an Independent Scien-tifie Review Committee (ISRC), to be appointed by the Administrator. This Committee was to complete a review of criteria documents and air quality standards by January 1,1980, and thereafter at five-year intervals, and recommend appropriate changes to the Administrator. Section 109(d)(2), 42 U.S.C. § 7409(d)(2). LIA argues that the Administrator contravened the Act by failing to submit the Lead Criteria Document and the lead standards to the ISRC for review in spite of the fact that Congress created the Committee in order to provide an opportunity for objective evaluation of the scientific issues raised by the task of setting air quality standards.

The Administrator chartered the ISRC as a subcommittee of EPA’s Science Advisory Board on January 18, 1978. By then EPA had already submitted three drafts of the Lead Criteria Document for review by the Lead Subcommittee of its Science Advisory Board and the public, the Subcommittee had substantially approved the third draft, and EPA, after considering the Subcommittee’s comments, had released the final Lead Criteria Document. We agree with the Administrator that in these circumstances there was little to gain from seeking further review of the Criteria Document by the newly established ISRC. The Science Advisory Board’s Lead Subcommittee was entirely composed of independent non-EPA experts, and its review of the various drafts of the Criteria Document was extremely thorough.121 In addition, nothing in Section 109(d) required the Administrator to resub*1172mit the Lead Criteria Document to the new committee. Accordingly, we reject LIA's claim that the Administrator violated the statute by not doing so.

D. Participation of Assistant Administrator Hawkins

LIA argues that the lead standards must be invalidated because of the participation of EPA Assistant Administrator David Hawkins in the rulemaking proceedings. Hawkins is head of EPA’s Office of Air, Noise, and Radiation. In that capacity he supervised development of the lead (and other) air quality standards and formally recommended adoption of the 1.5 ug Pb/m3 standard to the Administrator.122 Before he joined EPA Hawkins was a staff attorney with the Natural Resources Defense Council, Inc. (NRDC).123 LIA alleges that while at NRDC Hawkins “was very much involved in the lead standard issue as an attorney for NRDC,” and in this capacity “represented NRDC on the lead standard issue * * * before Congress and the National Air Quality Advisory Committee * * *."124 And it contends that because of this prior involvement Hawkins should have been disqualified from participating in the lead standards rulemaking proceedings “under the well-established rule that a government official may not participate in a matter in which he earlier participated in a representative or investigative capacity either within or without the agency.”125 Citing this court’s decision in Amos Treat & Co. v. SEC, 306 F.2d 260 (D.C.Cir.1962),126 LIA argues that Hawkins’ participation in the rulemaking proceedings violated its right to due process. In addition, LIA contends that Hawkins’ participation deprived the lead standards proceedings of the appearance of fairness which this court in the Amos Treat case described as essential to the administrative process. The only appropriate remedy, in LIA’s view, is for this court to vacate and remand the lead standards to EPA for further consideration without the participation of Assistant Administrator Hawkins.

Petitioner’s objection to Hawkins’ participation in the lead standards rulemaking faces a threshold obstacle which LIA must surmount before we can consider the merits of its claim. Section 307(d)(7)(B) of the Act, 42 U.S.C. § 7607(d)(7)(B), provides that “[o]nly an objection to a rule or procedure which was raised with reasonable specificity during the period for public comment (including any public hearing) may be raised during judicial review.” Exceptions to this requirement are permitted if the objection concerns an issue that was of central relevance to the outcome of the rule, and if it was impracticable to raise the objection during the comment period or if the grounds for the objection only arose after the comment period had expired. Id. LIA did not raise the issue of Hawkins’ disqualification until after the final regulations had been promulgated. It first brought the matter to the Administrator’s attention on December 8, 1978, two months after the final regulations were published in the Federal Register, when it raised the issue in its Petition for Reconsideration and Stay of the order adopting the final lead air quality standards.127

LIA presents a number of arguments in support of its view that its failure to present its objections to Hawkins’ participation during the public comment period as required by statute should not preclude this court from considering the merits of the claim. First, it denies it was aware of any grounds for Hawkins’ disqualification until November 27, 1978, when a newspaper reported that Hawkins was recusing himself from participating in an EPA decision concerning listing of arsenic as a hazardous *1173pollutant under Section 112 of the Act, 42 U.S.C. § 7412.128 Noting that the general rule governing disqualification requires such claims to be raised “as soon as practicable after a party has reasonable cause to believe that grounds for disqualification exist,” Marcus v. Director, Office of Wkrs’ Comp. Programs, 548 F.2d 1044, 1051 (D.C. Cir.1976), LIA argues that its conduct satisfied this test since its Petition for Reconsideration was filed two weeks after the newspaper report was published. In addition, it suggests that its challenge to Hawkins’ participation falls within the statutory exemption for objections that arose after the public comment period. Second, LIA contends that the issues raised by Hawkins’ participation in the proceedings are of sufficient importance to invoke this court’s inherent jurisdiction and responsibility “to decide whether there occurred here such an inroad upon the integrity of the decisional function of the independent agency as to require the court sua sponte to set aside the whole or any part of” the regulations. American Public Gas Ass’n v. FPC, 567 F.2d 1016, 1070 (D.C.Cir.1977), cert. denied, 435 U.S. 907, 98 S.Ct. 1456, 55 L.Ed.2d 499 (1978). Finally, LIA suggests that the timeliness requirement of Section 307(d)(7)(B) does not apply to its due process challenge to Hawkins’ participation in the proceedings because this is a claim that must be reviewed pursuant to Section 307(d)(9)(B), 42 U.S.C. § 760.7(d)(9)(B), which permits reversal of EPA actions found to be “contrary to constitutional right, power, privilege, or immunity.” It reads the decision in Porter v. Calif ano, 592 F.2d 770, 780 (5th Cir. 1979),129 construing an identical provision in the Administrative Procedure Act, 5 U.S.C. § 706(2)(B) (1976), as suggesting that this court should reach the merits of its objection to Hawkins’ participation.

Taking petitioner’s last argument first, we find LIA simply wrong in suggesting that Section 307(d)(9)(B), in allowing the court to set aside agency actions which violate constitutional rights, somehow releases it from the obligation to present its constitutional claims to the Agency in timely fashion, as specified by the Act. By the very terms of the statute Section 307(d)(7)(B)’s timeliness requirement applies to all objections, not just nonconstitutional challenges. Moreover, LIA’s interpretation of the statute would give parties to Clean Air Act proceedings a powerful weapon for delaying and sandbagging Agency action. They could simply refrain from presenting their constitutional objections to the Agency’s action until after the Agency had announced its final decision, then raise the issue before the reviewing court and obtain a reversal of the Agency’s decision, and thereby compel the Agency to institute new proceedings. It is difficult to believe that a Congress that has expressed concern about the need for expeditious attainment of the goals of the Clean Air Act130 would give parties affected by regulations issued under the Act such a potent weapon for delaying Agency action. Furthermore, LIA seriously misconstrues the decision in Porter v. Califano, supra, if it interprets it as suggesting that a litigant may present his constitutional objection to a reviewing court after failing to bring it to the Agency’s attention in timely fashion. The Fifth Circuit’s statement that the “intent of Congress * * * was that courts should make an independent assessment of a citizen’s claim of constitutional right when reviewing agency decision-making,” 592 F.2d at 780, merely acknowledges that a reviewing court owes no deference to the *1174agency’s pronouncement on a constitutional question. It in no way suggests that a litigant who neglected to present his constitutional claim to the administrative agency in timely fashion may not be precluded from raising it before the reviewing court.131 Accordingly, LIA’s objection to Assistant Administrator Hawkins’ participation in the lead standards rulemaking is properly before this court only if one of petitioner’s other two arguments succeeds in preserving the challenge.

In connection with LIA’s claim that it first had cause to believe Hawkins should have been disqualified after the final regulations had been published, it is important to note that LIA appears to advance two theories in support of its view that Hawkins’ participation in the rulemaking was improper. Both theories are founded on a suggestion that Hawkins may have been biased. First, LIA invokes the doctrine of conflict of interest, pointing out that NRDC was responsible for the litigation which forced the Administrator to list lead as a Section 108 pollutant, and that NRDC was a participant in the lead standards rulemak-ing.132 Under this theory the mere fact of Hawkins’ past position with NRDC — a participant in the proceedings — would allegedly justify the inference of bias. Second, LIA makes much of the allegation that Hawkins himself had been “very much involved” in the lead standards issue while he was employed at NRDC and had represented NRDC in the matter. While it is possible to construe this claim as further support for LIA’s conflict of interest charge, it also seems to suggest that LIA believes Hawkins’ prior involvement with the lead standards raises a prejudgment question. Bias theoretically could be demonstrated under this theory by pointing to specific conduct on Hawkins’ part that reveals he had made up his mind about the specific factual and legal issues raised by the standard-setting exercise before the rulemaking proceedings began. Although LIA makes little attempt to distinguish these theories, we believe they deserve separate consideration, and we conclude that only on the prejudgment claim can LIA surmount the threshold procedural barrier.

Accepting LIA’s claim that Hawkins played a major role in EPA’s decision-making, and assuming for the moment that LIA’s allegation of conflict of interest can be sustained, we are unable to perceive any basis for LIA’s suggestion that it could not have raised this issue with the Administrator during the public comment period. We must presume that LIA knew that NRDC was responsible for the litigation which led to the listing of lead as a Section 108 pollutant, if only because the preamble to the proposed lead standards specifically noted this fact.133 Nor can LIA claim ignorance about Hawkins’ involvement in the lead standards rulemaking at EPA. He was a member of the EPA panel that received testimony at the public hearings on the proposed standards, and he was specifically introduced as the Assistant Administrator for Air and Waste Management.134 Representatives of LIA, including its counsel, were present at these hearings. Finally, it is reasonable to assume that LIA or any other group that was involved in Clean Air *1175Act dealings with EPA knew that Hawkins had been employed by NRDC before he came to EPA. Thus LIA either knew or should have know about whatever possible conflict of interest problem may have been posed by Hawkins’ participation in these proceedings before the comment period expired. Yet at no time during the two years that this rulemaking took place did LIA object to, or even ask any questions about, Hawkins’ participation. If LIA really believed that Hawkins’ participation raised a conflict of interest question, it would have raised the issue at the time.135 Of course, it may be that LIA made a conscious tactical decision not to raise this question until after the rulemaking was completed, hoping to persuade the reviewing court to remand the standards to EPA and thereby delay the rulemaking. In any event, since LIA failed to present its conflict of interest claim to EPA in timely fashion, the statute precludes it from raising this issue before this court, and so we need not address the merits of LIA’s conflict of interest theory.136

To the extent that LIA’s objection to Hawkins’ participation in the lead standards rulemaking rests on a prejudgment theory, LIA’s claim that it only became aware of the grounds for his disqualification after the final regulations were adopted is slightly more plausible. Thus, while even this claim is difficult to believe,137 we have examined the merits of LIA’s prejudg*1176ment argument. We conclude that in the circumstances presented by this case Hawkins’ disqualification from the lead standards rulemaking is unwarranted.

First, nothing in the record indicates that Hawkins was involved with the particular issues raised in the lead standards rulemak-ing while he was at NRDC. In this connection it is important to spell out precisely what Hawkins’ activities concerning lead were at NRDC. In the affidavit he filed with this court Hawkins affirms that he “did not have any significant personal participation in NRDC’s efforts to have EPA list lead as a Section 108 pollutant, including its judicial action against EPA, or any other NRDC activity that concerned lead.”138 Hawkins did represent NRDC at a meeting of the National Air Quality Advisory Committee held on November 14, 1974, when the NRDC lawyer responsible for the lead issue was unable to attend.139 At this meeting he articulated NRDC’s previously stated position that the Administrator was required to list lead as a Section 108 pollutant.140 In addition,'Hawkins testified on behalf of NRDC at a Senate Subcommittee hearing in 1975, and in his testimony referred to lead as an example of EPA’s refusal to promulgate air quality standards for a pollutant even after the statutory criteria mandating a standard had been met.141 This reference to the lead listing controversy was a small section of a prepared statement he read at the hearing, most of which dealt with the unrelated issue of possible amendments to the Act that would preempt state authority.142 Hawkins resigned from NRDC before he joined EPA, and he never appeared in or in any way participated on NRDC’s behalf in the EPA proceedings to establish air quality standards for lead.

This description of Hawkins’ lead-related activities while at NRDC reveals two fatal deficiencies in LIA’s objection to his participation in the lead standards rulemaking. First, LIA’s contention that Hawkins was “very much involved in the lead standards issue as an attorney for NRDC,”143 is a gross exaggeration. His articulation of NRDC’s previously stated position on the lead listing controversy at two public forums hardly amounts to active involvement in this matter. Second, and more important, his testimony does not even come close to suggesting that Hawkins had prejudged the precise issues in this case because he never even adverted to them. Hawkins’ statements addressed only the question whether the Act required EPA to list lead as a pollutant under Section 108, an issue that was ultimately resolved by the NRDC litigation. This question is entirely separate from the issues that were raised in the lead standards rulemaking. The former involved a narrow question of statutory interpretation — whether or not the Administrator’s function to list pollutants under Section 108 is mandatory once the conditions specified by the Act are satisfied. See Natural Resources Defense Council, Inc. v. Train, supra, 411 F.Supp. at 867. There were no factual issues in this controversy, since EPA conceded that it had made the factual determinations required by Section 108. Id. The lead standards rulemaking, on the other hand, was concerned with deciding the level at which air quality stan*1177dards for lead should be set. While these proceedings also raised questions of statutory construction — to be sure, of different provisions of the Act — they primarily required decisions about policy and fact, most of which involved difficult scientific and technical questions144 As such, the issues and administrative process involved in listing lead as a Section 108 pollutant are quite separate and different from the issues and process involved in setting ambient air quality standards for lead under Section 109. And any views that Hawkins may have held or expressed concerning the lead listing controversy provide no evidence of his having prejudged, or even having any prior involvement with, the issues that confronted him in the lead standards rulemak-ing.145

LIA’s inability to present any evidence to support a claim that Hawkins may have prejudged the precise factual and legal issues raised by the lead standards proceedings distinguishes this case from the “prior involvement” decisions of, this and other circuits on which LIA relies. In Amos Treat & Co. v. SEC, supra, we ruled that it was improper for a member of the Commission to participate in an adjudicatory proceeding when, before he became a Commissioner, he had, as director of the Commission’s Division of Corporate Finance, supervised the initiation and conduct of the investigation which resulted in the proceeding.146 The court held that it was a violation of the notion of “fair play” inherent in due process for “a member of an investigative or prosecuting staff [to] initiate an investigation, weigh its results, perhaps then recommend the filing of charges, and thereafter * * * participate in adjudicatory proceedings * * 306 F.2d at 266. And in American Cyanamid Co. v. FTC, 363 F.2d 757 (6th Cir. 1966), the Sixth Circuit disqualified Chairman Dixon of the FTC from an adjudicatory proceeding because- he had previously served as Chief Counsel and Staff Director of a Senate Subcommittee which had conducted an investigation into many of the same legal and factual issues that were before the Commission. Significantly, the court’s decision in that case was based on the depth of the Subcommittee’s investigation of the precise factual issues which were presented in the proceedings before the Commission, the very active role Mr. Dixon had played in the conduct of the investigation, and the uncon-troverted evidence in the record which indicated that he had formed conclusions about the particular factual issues that were before the Commission.147

In both these cases the intimate involvement of the decisionmakers in the investigation of the same factual and legal issues they were then called upon to adjudicate made it inevitable for “a disinterested observer” to conclude that they had “in some measure adjudged the facts as well as the law * * * in advance of hearing [the *1178cases].” Cinderella Career & Finishing Schools, Inc. v. FTC, 425 F.2d 583, 591 (D.C. Cir.1970) (emphasis added), quoting Gilligan, Will & Co. v. SEC, 267 F.2d 461, 469 (2d Cir.), cert. denied, 361 U.S. 896, 80 S.Ct. 200, 4 L.Ed.2d 152 (1959).148 In contrast, Hawkins’ only involvement with the lead listing controversy was his appearance as a spokesperson on a public issue at a' public forum, and even this involvement appears to have been partly fortuitous. Nothing in his statements suggests that he had prejudged any of the legal questions that were presented by the lead standards rulemaking. More important, insofar as both Amos Treat and American Cyanamid require a showing that the decisionmaker had formed conclusions about the particular facts of the case, nothing in the record indicates that Hawkins had taken a position apparently inconsistent with an ability to judge the factual issues that had to be resolved in setting air quality standards for lead fairly. And there is most certainly nothing in the record which suggests that Hawkins had prejudged the issue of the level at which the air quality standards for lead should be set. Accordingly, we find no basis for requiring Hawkins’ disqualification from the lead standards proceedings on prejudgment grounds.149

To be sure, by publicly articulating NRDC’s position that the Administrator was required to list lead as a Section 108 pollutant Hawkins may have advocated promulgation of air quality standards for lead, and such advocacy might appear to make his participation in the subsequent EPA proceeding to set the standards improper. The short answer to this view is that, according to the cases, a decisionmaker “is [not] disqualified simply because he has taken a position, even in public, on a policy issue related to a dispute, in the absence of a showing that he is not ‘capable of judging a particular controversy fairly on the basis of its own circumstances.’ ” Hortonville Joint School Dist. No. 1 v. Hortonville Education Ass’n, 426 U.S. 482, 493, 96 S.Ct. 2308, 2314, 49 L.Ed.2d 1 (1976), quoting United States v. Morgan, 313 U.S. 409, 421, 61 S.Ct. 999, 1004, 85 L.Ed. 1429 (1941). This test has been applied in deciding disqualification questions involving judges as well as administrative agency decisionmak-ers participating in adjudicatory proceedings. See, e.g. Laird v. Tatum, 409 U.S. 824, 831-836, 93 S.Ct. 7, 11-14, 34 L.Ed.2d 50 (memorandum of Rehnquist, J.); Anto-nello v. Wunsch, 500 F.2d 1260, 1262 (10th Cir. 1974); Goodpasture v. TVA, 434 F.2d 760, 765 (6th Cir. 1970); Knoll v. Socony Mobil Oil Co., 369 F.2d 425, 430 (10th Cir. 1966) (judges); FTC v. Cement Institute, 333 U.S. 683, 702-703, 68 S.Ct. 793, 804, 92 L.Ed. 1010 (1948); Carolina Environmental Study Group v. United States, 510 F.2d 796, 801 (D.C.Cir.1975); Corning Glass Works v. FTC, 509 F.2d 293, 303-304 (7th Cir. 1975); American Cyanamid Co. v. FTC, supra, 363 F.2d at 765-767 (agency decisionmakers).

*1179The rule could hardly be otherwise, particularly with respect to administrative agencies which are created for the specific purpose of accomplishing certain tasks. Agency decisionmakers are appointed precisely to implement statutory programs, and so inevitably have some policy'preconceptions. See Carolina Environmental Study Group v. United States, supra, 510 F.2d at 801. As Professor Davis has pointed out:

* * * A Trade Commissioner should not be neutral on anti-monopoly policies, and a Securities and Exchange Commissioner should not be apathetic about the need for governmental restrictions. * *

* * * The theoretically ideal administrator is one whose broad point of view is in general agreement with the policies he administers * * * .C150]

To be sure, a different question may be presented if it can be shown that an agency decisionmaker has exhibited the type of single-minded commitment to a particular position that makes him or her totally incapable of giving fair consideration to the issues that are presented for decision. This is not, however, such a case. Nothing in the record suggests that Assistant Administrator Hawkins was incapable of considering the issues raised by the lead standards proceedings fairly.

Thus far we have examined Hawkins’ participation in the lead standards proceedings under the prejudgment test that has been applied in adjudicatory proceedings. It is, however, beyond dispute that due process may impose different procedural requirements in an adjudication than are imposed in a rulemaking. United States v. Florida East Coast R. Co., 410 U.S. 224, 244-245, 93 S.Ct. 810, 820-21, 35 L.Ed.2d 223 (1973); United States v. Allegheny-Ludlum Steel Corp., 406 U.S. 742, 92 S.Ct. 1941, 32 L.Ed.2d 453 (1976). Our opinion in Amos Treat explained that “[j]ust exactly how the concept of ‘due process’ is to be applied will vary with the type of proceeding involved * * * Amos Treat & Co. v. SEC, supra, 306 F.2d at 263, and it discussed at length the reasons why the result in that case was mandated by the adjudicatory nature of the SEC proceedings. Id. at 263-264. See also American Cyanamid Co. v. FTC, supra, 363 F.2d at 766-767. LIA is unable to point to any cases involving rulemaking proceedings in which a court has held that a decisionmaker’s prior involvement with the issues presented in the proceedings, whether within or without the agency, is sufficient ground for requiring his or her disqualification.151 To the contrary, a division of this court recently held that Cinderella’s prejudgment test does not apply to rulemaking proceedings.152 Ass’n of Nat’l Advertisers, Inc. v. FTC, 627 F.2d 1151 (D.C.Cir.1979). Judge Tamm’s opinion for the court contains an illuminating discussion of important distinctions between rulemaking and adjudication, and the consequences of these differences for standards of disqualification, which we will not repeat here. See id., *1180627 F.2d at 1162-1165, at 1170-1174. See also id., 627 F.2d at 1175-1181 (Leventhal, J., concurring). And under the prejudgment test for rulemaking announced in that case — a clear and convincing showing of an unalterably closed mind on a matter critical to disposition of the proceeding — there can be no question but that Hawkins’ disqualification from the lead standards rulemaking is unwarranted.153

IX. THE SECONDARY AIR QUALITY STANDARD

The final challenge to the Administrator’s actions that we must consider is LIA’s objection to his decision to promulgate a national secondary ambient air quality standard for lead of 1.5 ug Pb/m3, the same level as the primary standard. The Administrator explained that this decision was based on his conclusion that the evidence on the “welfare effects” of lead exposure did not justify promulgation of a secondary standard more stringent than the primary standard. LIA argues that the Administrator had no authority to adopt the secondary standard without making supporting findings showing that the standard is necessary to protect the public welfare. It contends that this court’s decision in Kennecott Copper Corp. v. EPA, 462 F.2d 846 (D.C.Cir. 1972), mandates remand of the secondary standard. In that case we remanded the national secondary air quality standard for sulphur dioxide because the Administrator had failed to “enlighten the court as to the basis on which he reached the * * * standard from the material in the Criteria [Document].” Id. at 850. LIA argues that the Administrator should have considered whether the welfare effects of lead exposure would permit him to set a secondary air quality standard higher than the primary standard. Pointing out that the primary standards are supposed to protect public health and that they are based on a 90-day averaging period, LIA argues that it follows that primary standards need be met only in inhabited areas, whereas the secondary standards must also be met in uninhabited areas. Consequently, LIA contends, the Administrator’s failure to consider the possibility of setting a higher secondary standard will impose an additional burden on industrial sources located in uninhabited areas which may not be justified by the requirements of protecting the public welfare.

LIA’s complaint is based on a misconception about the reach of the primary standard. As EPA notes, the primary standard must be met in all parts of the country, whether inhabited or uninhabited.154 Thus by setting the secondary standard at the same level as the primary standard, the Administrator imposed no additional burdens on the industry, and he properly concentrated his attention on whether the welfare effects of lead exposure justified promulgation of a more stringent secondary standard. Our decision in Kennecott Copper, on which LIA relies, involved an attempt by the Administrator to set a secondary standard which was more stringent than the primary standard without explaining the basis for this decision, see 36 Fed. Reg. 8187, and it is therefore inapposite to the instant case. Furthermore, LIA did not object to the Administrator’s proposal to set the secondary standard at the same level as the primary standard either in the comments that it filed on the proposed standards or at any other time during the public comment period. Indeed, so far as we have been able to determine, none of the participants in the rulemaking proceedings commented on the proposal. Accordingly, LIA *1181is precluded from raising this objection by the timeliness requirement of the Act.155

X. LIA’S MOTION TO SUPPLEMENT THE ADMINISTRATIVE RECORD

In addition to reviewing petitioners’ substantive and procedural challenges to the air quality standards for lead promulgated by EPA, we are also asked to rule on a motion filed by petitioner LIA for leave to supplement the index to the administrative record compiled by the Agency. LIA seeks to include in the record 38 documents — 37 of them internal EPA memoranda and the other a letter from a third party — which it claims bear directly on the questions of the fairness and rationality of the Agency’s de-cisionmaking process and procedures.156 The motion raises an issue of first impression concerning the proper scope of an administrative record assembled under Section 307(d) of the Act.

For many years courts reviewing agency decisions have struggled with the problem of large and unwieldly administrative records, particularly in reviewing informal rulemaking. More often than not the agencies did not begin to assemble the administrative record until after the regulations were challenged, with the result that courts were forced to review “historical” records— consisting of after-the-fact attempts to reconstruct the agency’s decisionmaking process. A major defect of this approach was that the court was often unable to determine which of the large number of documents that were dumped into the record played a significant role in the agency’s decision. Moreover, the agency could attempt to shore up inadequately justified positions by adding post hoc rationalizations to the record. Finally, participants in the rulemaking were not afforded an opportunity to comment on the materials that the agency considered relevant to the decision. In 1975 an EPA attorney published an article detailing these and other shortcomings of the “historical” approach to compiling the record. Pederson, Formal Records and Informal Rulemaking, 85 Yale L.J. 38 (1975). He recommended that this approach be replaced by a “procedural” approach to recordmaking, whereby the agency would compile the record as the rulemak-ing progressed and the record would be closed when the final rule was promulgated. This record would then be the exclusive record for the agency’s decision and judicial review. Id. at 78-82. Congress, in enacting Section 307(d), adopted these recommendations,157 and required EPA to include in the record “all data, information, and documents” on which the rule relies.158 In *1182this case EPA compiled a record in the manner required by the statute, and certified the index to the record.

After the final standards were promulgated LIA filed a request under the Freedom of Information Act (FOIA), 5 U.S.C. § 552 (1976), for all documents in EPA’s possession relating to the lead standards. After reviewing the documents it received, LIA concluded that it wished to include 38 documents in the administrative record, and it filed a request with EPA to this end. EPA denied the request, and thereafter LIA filed the instant, motion. The documents involved can be divided into three groups. The first group, Nos. 6, 9, 12, 15, 21-26, 33-38 (the Hawkins documents) are all supposed to show the role played by EPA Assistant Administrator Hawkins in the lead standards rulemaking. LIA argues *1183that they are relevant to its claim that Hawkins should have been disqualified from participating in the rulemaking. The second set of documents, Nos. 8, 16-20, BOSS (the economic evaluation documents), discuss the economic effects of the proposed standards. LIA contends that these documents show that EPA examined the economic impact of the lead standards, and it maintains that they should be included in the record because they are relevant to its arguments concerning the economic effects of the standards on the lead industry. The third set of documents, Nos. 1-5, 7, 10, 11, 13, 14, 27-29 (the staff opinions) are intra-agency memoranda159 which LIA claims bear directly on the questions of the fairness and rationality of the Agency’s deci-sionmaking process.

LIA rightly points out that the legislative history of Section 307(d) cautions the Agency against attempting to create a one-sided record by excluding from it material unfavorable to the Agency’s position. H.R.Rep. No.95-294, supra, at 319-320. And it notes, again correctly, that the legislative history acknowledges that parties have used the FOIA as an informal discovery device in rulemaking proceedings. Id. at 320. LIA then argues that its attempt to supplement the record with documents uncovered through a FOIA request is precisely what Congress intended should happen.

LIA’s reasoning is correct, so far as it goes, but it does not go far enough. Its FOIA request was not filed until after the final rule was promulgated, and by this time the record was closed. Nothing in the statute or its legislative history indicates that a party or the agency may reopen the record by placing additional materials (other than those required by the statute and wrongfully omitted by EPA) in the docket after promulgation of the rule. See American Petroleum Institute v. Costle, 609 F.2d 20, 22-23 (D.C.Cir.1979). The passage in the legislative history of Section 307(d) on which LIA relies merely directs EPA to coordinate its procedures for dealing with FOIA requests with the recordmaking provisions of the statute. H.R.Rep.No.95-294, supra, at 320. And it suggests that LIA should have filed its FOIA request during the interval between publication of the proposed and the final rules. Not having done so, it cannot now attempt to supplement the administrative record on appeal. Section 307(d)(7)(A), 42 U.S.C. § 7607(d)(7)(A).160

The Hawkins documents may, however present a special case. Since LIA claims that it only became aware of the grounds for Hawkins’ disqualification after the final regulations were promulgated, it obviously could not have filed a FOIA request for these documents before then. EPA agrees with LIA that these documents should be before the court, and it proposed a solution whereby these documents would be lodged with the court as part of a supplemental court record, separate from the administrative record. In this manner the documents would be available to the court in passing on the issue of Hawkins’ disqualification, and at the same time the recordmaking scheme established by the Act would be preserved. LIA has no objection to this proposal. Although we see no need for creating a separate record, we agree that the Hawkins documents are properly before the court for the sole purpose of ruling on LIA’s challenge to Hawkins’ participation in the rulemaking proceedings.161 Accord*1184ingly, we affirm EPA’s decision to exclude all 38 documents from the administrative record and, in addition, direct that Documents 6, 9, 12, 15, 21-26, 33-38 (the Hawkins documents) be lodged with the court.162

XI. CONCLUSION

The national ambient air quality standards for lead were the culmination of a process of rigorous scientific and public review which permitted a thorough ventilation of the complex scientific and technical issues presented by this rulemaking proceeding. Interested parties were allowed a number of opportunities to participate in exploration and resolution of the issues raised by the standard-setting exercise. EPA, and ultimately the public whose health these air quality standards protect, have benefitted from their contribution. To be sure, even the experts did not always agree about the answers to the questions that were raised. Indeed, they did not always agree on what the relevant questions were. These disagreements underscore the novelty and complexity of the issues that had to be resolved, and both the EPA and the participants in the rulemaking proceeding deserve to be commended for the diligence with which they approached the task of coming to grips with these difficult issues.

We have accorded these cases the most careful consideration, combining as we must careful scrutiny of the evidence in the record with deference to the Administrator’s judgments. We conclude that in this rulemaking proceeding the Administrator complied with the substantive and procedural requirements of the Act, and that his decisions are both adequately explained and amply supported by evidence in the record, Accordingly, we reject petitioners’ claims of error. The regulations under review herein are

Affirmed.

*462Justice Scalia

delivered the opinion of the Court.

These cases present the following questions: (1) Whether § 109(b)(1) of the Clean Air Act (CAA) delegates legislative power to the Administrator of the Environmental Protection Agency (EPA). (2) Whether the Administrator may consider the costs of implementation in setting national ambient air quality standards (NAAQS) under § 109(b)(1). (3) Whether the Court of Appeals had jurisdiction to review the EPA’s interpretation of Part D of Title I of the CAA, 42 U. S. C. §§ 7501-7515, with respect to implementing the revised ozone NAAQS. (4) If so, whether the EPA’s interpretation of that part was permissible.

I

. Section 109(a) of the CAA, as added, 84 Stat. 1679, and amended, 42 U. S. C. § 7409(a), requires the Administrator of the EPA to promulgate NAAQS for each air pollutant for which “air quality criteria” have been issued under § 108, 42 U. S. C. § 7408. Once a NAAQS has been promulgated, the Administrator must review the standard (and the criteria *463on which it is based) “at five-year intervals” and make “such revisions ... as may be appropriate.” CAA § 109(d)(1), 42 U. S. C. § 7409(d)(1). These cases arose when, on July 18, 1997, the Administrator revised the NAAQS for particulate matter and ozone. See NAAQS for Particulate Matter, 62 Fed. Reg. 38652 (codified in 40 CFR §50.7 (1999)); NAAQS for Ozone, id., at 38856 (codified in 40 CFR §§50.9, 50.10 (1999)). American Trucking Associations, Inc., and its corespondents in No. 99-1257 — which include, in addition to other private companies, the States of Michigan, Ohio, and West Virginia — challenged the new standards in the Court of Appeals for the District of Columbia Circuit, pursuant to 42 U. S. C. § 7607(b)(1).

The District of Columbia Circuit accepted some of the challenges and rejected others. It agreed with the No. 99-1257 respondents (hereinafter respondents) that § 109(b)(1) delegated legislative power to the Administrator in contravention of the United States Constitution, Art. I, § 1, because it found that the EPA had interpreted the statute to provide no “intelligible principle” to guide the agency’s exercise of authority. American Trucking Assns., Inc. v. EPA, 175 F. 3d 1027, 1034 (1999). The court thought, however, that the EPA could perhaps avoid the unconstitutional delegation by adopting a restrictive construction of § 109(b)(1), so instead of declaring the section unconstitutional the court remanded the NAAQS to the agency. Id., at 1038. (On this delegation point, Judge Tatel dissented, finding the statute constitutional as written. Id., at 1057.) On the second issue that the Court of Appeals addressed, it unanimously rejected respondents’ argument that the court should depart from the rule of Lead Industries Assn., Inc. v. EPA, 647 F. 2d 1130, 1148 (CADC 1980), that the EPA may not consider the cost of implementing a NAAQS in setting the initial standard. It also rejected respondents’ argument that the implementation provisions for ozone found in Part D, Sub-part 2, of Title I of the CAA, 42 U. S. C. §§ 7511-7511f, were *464so tied to the existing ozone standard that the EPA lacked the power to revise the standard. The court held that although Subpart 2 constrained the agency’s method of implementing the new standard, 175 F. 3d, at 1050, it did not prevent the EPA from revising the standard and designating areas of the country as “nonattainment areas,” see 42 U. S. C. § 7407(d)(1), by reference to it, 175 F. 3d, at 1047-1048. On the EPA’s petition for rehearing, the panel adhered to its position on these points, and unanimously rejected the EPA’s new argument that the court lacked jurisdiction to reach the implementation question because there had been no “final” implementation action. American Trucking Assns., Inc. v. EPA, 195 F. 3d 4 (CADC 1999). The Court of Appeals denied the EPA’s suggestion for rehearing en banc, with five judges dissenting. Id., at 13.

The Administrator and the EPA petitioned this Court for review of the first, third, and fourth questions described in the first paragraph of this opinion. Respondents conditionally cross-petitioned for review of the second question. We granted certiorari on both petitions, 529 U. S. 1129 (2000); 530 U. S. 1202 (2000), and scheduled the cases for argument in tandem. We have now consolidated the cases for purposes of decision.

II

In Lead Industries Assn., Inc. v. EPA, supra, at 1148, the District of Columbia Circuit held that “economic considerations [may] play no part in the promulgation of ambient air quality standards under Section 109” of the CAA. In the present cases, the court adhered to that holding, 175 F. 3d, at 1040-1041, as it had done on many other occasions. See, e. g., American Lung Assn. v. EPA, 134 F. 3d 388, 389 (1998); NRDC v. Administrator, EPA, 902 F. 2d 962, 973 (1990), vacated in part on other grounds, NRDC v. EPA, 921 F. 2d 326 (CADC 1991); American Petroleum Institute v. Costle, 665 F. 2d 1176, 1185 (1981). Respondents argue that these *465decisions are incorrect. We disagree; and since the first step in assessing whether a statute delegates legislative power is to determine what authority the statute confers, we address that issue of interpretation first and reach respondents’ constitutional arguments in Part III, infra.

Section 109(b)(1) instructs the EPA to set primary ambient air quality standards “the attainment and maintenance of which ... are requisite to protect the public health” with “an adequate margin of safety.” 42 U. S. C. § 7409(b)(1). Were it not for the hundreds of pages of briefing respondents have submitted on the issue, one would have thought it fairly clear that this text does not permit the EPA to consider costs in setting the standards. The language, as one scholar has noted, “is absolute.” D. Currie, Air Pollution: Federal Law and Analysis 4-15 (1981). The EPA, “based on” the information about health effects contained in the technical “criteria” documents compiled under § 108(a)(2), 42 U. S. C. § 7408(a)(2), is to identify the maximum airborne concentration of a pollutant that the public health can tolerate, decrease the concentration to provide an “adequate” margin of safety, and set the standard at that level. Nowhere are the costs of achieving such a standard made part of that initial calculation.

Against this most natural of readings, respondents make a lengthy, spirited, but ultimately unsuccessful attack. They begin with the object of § 109(b)(l)’s focus, the “public health.” When the term first appeared in federal clean air legislation — in the Act of July 14, 1955 (1955 Act), 69 Stat. 322, which expressed “recognition of the dangers to the public health” from air pollution — its ordinary meaning was “[t]he health of the community.” Webster’s New International Dictionary 2005 (2d ed. 1950). Respondents argue, however, that § 109(b)(1), as added by the Clean Air Amendments of 1970,84 Stat. 1676, meant to use the term’s secondary meaning: “[t]he ways and means of conserving the health *466of the members of a community, as by preventive medicine, organized care of the sick, etc.” Ibid. Words that can have more than one meaning are given content, however, by their surroundings, FDA v. Brown & Williamson Tobacco Corp., 529 U. S. 120, 182-183 (2000); Jones v. United States, 527 U. S. 373, 389 (1999), and in the context of § 109(b)(1) this second definition makes no sense. Congress could not have meant to instruct the Administrator to set NAAQS at a level “requisite to protect” “the art and science dealing with the protection and improvement of community health.” Webster’s Third New International Dictionary 1836 (1981). We therefore revert to the primary definition of the term: the health of the public.

Even so, respondents argue, many more factors than air pollution affect public health. In particular, the economic cost of implementing a very stringent standard might produce health losses sufficient to offset the health gains achieved in cleaning the air — for example, by closing down whole industries and thereby impoverishing the workers and consumers dependent upon those industries. That is unquestionably true, and Congress was unquestionably aware of it. Thus, Congress had commissioned in the Air Quality Act of 1967 (1967 Act) “a detailed estimate of the cost of carrying out the provisions of this Act; a comprehensive study of the cost of program implementation by affected units of government; and a comprehensive study of the economic impact of air quality standards on the Nations industries, communities, and other contributing sources of pollution.” §2, 81 Stat. 505. The 1970 Congress, armed with the results of this study, see The Cost of Clean Air, S. Doc. No. 91-40 (1969) (publishing the results of the study), not only anticipated that compliance costs could injure the public health, but provided for that precise exigency. Section 110(f)(1) of the CAA permitted the Administrator to waive the compliance deadline for stationary sources if, inter *467alia, sufficient control measures were simply unavailable and “the continued operation of such sources is essential... to the public health or welfare.” 84 Stat. 1683 (emphasis added). Other provisions explicitly permitted or required economic costs to be taken into account in implementing the air quality standards. Section 111(b)(1)(B), for example, commanded the Administrator to set “standards of performance” for certain new sources of emissions that as specified in § 111(a)(1) were to “reflec[t] the degree of emission limitation achievable through the application of the best system of emission reduction which (taking into account the cost of achieving such reduction) the Administrator determines has been adequately demonstrated.” Section 202(a)(2) prescribed that emissions standards for automobiles could take effect only “after such period as the Administrator finds necessary to permit the development and application of the requisite technology, giving appropriate consideration to the cost of compliance within such period.” 84 Stat. 1690. See also § 202(b)(5)(C) (similar limitation for interim standards); § 211(c)(2) (similar limitation for fuel additives); § 231(b) (similar limitation for implementation of aircraft emission standards). Subsequent amendments to the CAA have added many more provisions directing, in explicit language, that the Administrator consider costs in performing various duties. See, e. g., 42 U. S. C. § 7545(k)(1) (reformulate gasoline to “require the greatest reduction in emissions ... taking into consideration the cost of achieving such emissions reductions”); § 7547(a)(3) (emission reduction for nonroad vehicles to be set “giving appropriate consideration to the cost” of the standards). We have therefore refused to find implicit in ambiguous sections of the CAA an authorization to consider costs that has elsewhere, and so often, been expressly granted. See Union Elec. Co. v. EPA, 427 U. S. 246, 257, and n. 5 (1976). Cf. General Motors Corp. v. United States, 496 U. S. 530, 538, 541 (1990) *468(refusing to infer in certain provisions of the CAA deadlines and enforcement limitations that had been expressly imposed elsewhere).

Accordingly, to prevail in their present challenge, respondents must show a textual commitment of authority to the EPA to consider costs in setting NAAQS under § 109(b)(1). And because § 109(b)(1) and the NAAQS for which it provides are the engine that drives nearly all of Title I of the CAA, 42 U. S. C. §§7401-7515, that textual commitment must be a clear one. Congress, we have held, does not alter the fundamental details of a regulatory scheme in vague terms or ancillary provisions — it does not, one might say, hide elephants in mouseholes. See MCI Telecommunications Corp. v. American Telephone & Telegraph Co., 512 U. S. 218, 231 (1994); FDA v. Brown & Williamson Tobacco Corp., supra, at 159-160. Respondents’ textual arguments ultimately founder upon this principle.

Their first claim is that §109(b)(l)’s terms “adequate margin” and “requisite” leave room to pad health effects with cost concerns. Just as we found it “highly unlikely that Congress would leave the determination of whether an industry will be entirely, or even substantially, rate-regulated to agency discretion — and even more unlikely that it would achieve that through such a subtle device as permission to ‘modify’ rate-filing requirements,” MCI Telecommunications Corp. v. American Telephone & Telegraph Co., supra, at 231, so also we find it implausible that Congress would give to the EPA through these modest words the power to determine whether implementation costs should moderate national air quality standards. Accord, Christensen v. Harris County, 529 U. S. 576, 590, n. (2000) (Scalia, J., concurring in part and concurring in judgment) (“The implausibility of Congress’s leaving a highly significant issue unaddressed (and thus ‘delegating’ its resolution to the administering agency) is assuredly one of the factors *469to be considered in determining whether there is ambiguity” (emphasis deleted)).1

The same defect inheres in respondents’ next two arguments: that while the Administrator’s judgment about what is requisite to protect the public health must be “based on [the] criteria” documents developed under § 108(a)(2), see § 109(b)(1), it need not be based solely on those criteria; and that those criteria themselves, while they must include “effects on public health or welfare which may be expected from the presence of such pollutant in the ambient air,” are not necessarily limited to those effects. Even if we were to concede those premises, we still would not conclude' that one of the unenumerated factors that the agency can consider in developing and applying the criteria is cost of implementation. That factor is both so indirectly related to public health and so full of potential for canceling the conclusions drawn from direct health effects that it would surely have been expressly mentioned in §§ 108 and 109 had Congress meant it to be considered. Yet while those provisions describe in detail how the health effects of pollutants in the ambient air are to be calculated and given effect, see § 108(a)(2), they say not a word about costs.

Respondents point, finally, to a number of provisions in the CAA that do require attainment cost data to be generated. Section 108(b)(1), for example, instructs the Administrator to “issue to the States,” simultaneously with the criteria documents, “information on air pollution control techniques, which information shall include data relating to the cost of installation and operation.” 42 U. S. C. § 7408(b)(1). And *470§ 109(d)(2)(C)(iv) requires the Clean Air Scientific Advisory Committee to “advise the Administrator of any adverse public health, welfare, social, economic, or energy effects which may result from various strategies for attainment and maintenance” of NAAQS.2 42 U. S. C. § 7409(d)(2)(C)(iv). Respondents argue that these provisions make no sense unless costs are to be considered in setting the NAAQS. That is not so. These provisions enable the Administrator to assist the States in carrying out their statutory role as primary implementers of the NAAQS. It is to the States that the CAA assigns initial and primary responsibility for deciding what emissions reductions will be required from which sources. See 42 U. S. C. §§ 7407(a), 7410 (giving States the duty of developing implementation plans). It would be impossible to perform that task intelligently without considering which abatement technologies are most efficient, and most economically feasible — which is why we have said that “the most important forum for consideration of claims of economic and technological infeasibility is before the state agency formulating the implementation plan,” Union Elec. Co. v. EPA, 427 U. S., at 266. Thus, federal clean air legislation has, from the very beginning, directed federal agencies to develop and transmit implementation data, including cost data, to the States. See 1955 Act, *471§2(b), 69 Stat. 322; Clean Air Act of 1963, amending §§3(a), (b) of the CAA, 77 Stat. 394; 1967 Act, §§ 103(a)-(d), 104, 107(c), 81 Stat. 486-488. That Congress chose to carry forward this research program to assist States in choosing the means through which they would implement the standards is perfectly sensible, and has no bearing upon whether cost considerations are to be taken into account in formulating the standards.3

It should be clear from what we have said that the canon requiring texts to be so construed as to avoid serious constitutional problems has no application here. No matter how severe the constitutional doubt, courts may choose only between reasonably available interpretations of a text. See, e. g., Miller v. French, 530 U. S. 327, 341 (2000); Pennsylvania Dept. of Corrections v. Yeskey, 524 U. S. 206, 212 (1998). The text of § 109(b), interpreted in its statutory and historical context and with appreciation for its importance to the CAA as a whole, unambiguously bars cost considerations from the NAAQS-setting process, and thus ends the matter for us as well as the EPA.4 We therefore affirm the judgment of the Court of Appeals on this point.

*472III

Section 109(b)(1) of the CAA instructs the EPA to set “ambient air quality standards the attainment and maintenance of which in the judgment of the Administrator, based on [the] criteria [documents of § 108] and allowing an adequate margin of safety, are requisite to protect the public health.” 42 U. S. C. § 7409(b)(1). The Court of Appeals held that this section as interpreted by the Administrator did not provide an “intelligible principle” to guide the EPA’s exercise of authority in setting NAAQS. “[The] EPA,” it said, “lack[ed] any determinate criteria for drawing lines. It has failed to state intelligibly how much is too much.” 175 F. 3d, at 1034. The court hence found that the EPA’s interpretation (but not the statute itself) violated the non-delegation doctrine. Id., at 1038. We disagree.

In a delegation challenge, the constitutional question is whether the statute has delegated legislative power to the agency. Article I, § 1, of the Constitution vests “[a]ll legislative Powers herein granted ... in a Congress of the United States.” This text permits no delegation of those powers, Loving v. United States, 517 U. S. 748, 771 (1996); see id., at 776-777 (Scalia, J., concurring in part and concurring in judgment), and so we repeatedly have said that when Congress confers decisionmaking authority upon agencies Congress must “lay down by legislative act an intelligible principle to which the person or body authorized to [act] is directed to conform.” J. W. Hampton, Jr., & Co. v. United States, 276 U. S. 394, 409 (1928). We have never suggested that an agency can cure an unlawful delegation of legislative power by adopting in its discretion a limiting construction of the statute. Both Fahey v. Mallonee, 332 U. S. 245, 252-253 (1947), and Lichter v. United States, 334 U. S. 742, 783 (1948), mention agency regulations in the course of their nondelegation discussions, but Lichter did so because a subsequent Congress had incorporated the regulations into a revised version of the statute, ibid., and Fahey because the custom*473ary practices in the area, implicitly incorporated into the statute, were reflected in the regulations, 832 U. S., at 250. The idea that an agency can cure an unconstitutionally stand-ardless delegation of power by declining to exercise some of that power seems to us internally contradictory. The very choice of which portion of the power to exercise — that is to say, the prescription of the standard that Congress had omitted — would itself be an exercise of the forbidden legislative authority. Whether the statute delegates legislative power is a question for the courts, and an agency’s voluntary self-denial has no bearing upon the answer.

We agree with the Solicitor General that the text of § 109(b)(1) of the CAA at a minimum requires that “[flor a discrete set of pollutants and based on published air quality criteria that reflect the latest scientific knowledge, [the] EPA must establish uniform national standards at a level that is requisite to protect public health from the adverse effects of the pollutant in the ambient air.” Tr. of Oral Arg. ih-No. 99-1257, p. 5. Requisite, in turn, “mean[s] sufficient, but not more than necessary.” Id., at 7. These limits on the EPA’s discretion are strikingly similar to the ones we approved in Touby v. United States, 500 U. S. 160 (1991), which permitted the Attorney General to designate a drug as a controlled substance for purposes of criminal drug enforcement if doing so was “ ‘necessary to avoid an imminent hazard to the public safety.’” Id., at 163. They also resemble the Occupational Safety and Health Act of 1970 provision requiring the agency to “ ‘set the standard which most adequately assures, to the extent feasible, on the basis of the best available evidence, that no employee will suffer any impairment of health’ ” — which the Court upheld in Industrial Union Deyt., AFL-GIO v. American Petroleum Institute, 448 U. S. 607, 646 (1980), and which even then-JuSTlCE Rehnquist, who alone in that case thought the statute violated the nondelegation doctrine, see id., at 671 (opinion concurring in judgment), would have upheld if, like the statute *474here, it did not permit economic costs to be considered. See American Textile Mfrs. Institute, Inc. v. Donovan, 452 U. S. 490, 545 (1981) (Rehnquist, J., dissenting).

The scope of discretion § 109(b)(1) allows is in fact well within the outer limits of our nondelegation precedents. In the. history of the Court we have found the requisite “intelligible principle” lacking in only two statutes, one of which provided literally no guidance for the exercise of discretion, and the other of which conferred authority to regulate the entire economy on the basis of no more precise a standard than stimulating the economy by assuring “fair competition.” See Panama Refining Co. v. Ryan, 293 U. S. 388 (1935); A. L. A. Schechter Poultry Corp. v. United States, 295 U. S. 495 (1935). We have, on the other hand, upheld the validity of § 11(b)(2) of the Public Utility Holding Company Act of 1935, 49 Stat. 821, which gave the Securities and Exchange Commission authority to modify the structure of holding company systems so as to ensure that they are not “unduly or unnecessarily complicate^]” and do not “unfairly or inequitably distribute voting power among security holders.” American Power & Light Co. v. SEC, 329 U. S. 90, 104 (1946). We have approved the wartime conferral of agency power to fix the prices of commodities at a level that “ ‘will be generally fair and equitable and will effectuate the [in some respects conflicting] purposes of th[e] Act.’” Yakus v. United States, 321 U. S. 414, 420, 423-426 (1944). And we have found an “intelligible principle” in various statutes authorizing regulation in the “public interest,” See, e. g., National Broadcasting Co. v. United States, 319 U. S. 190, 225-226 (1943) (Federal Communications Commission’s power to regulate airwaves); New York Central Securities Corp. v. United States, 287 U. S. 12, 24-25 (1932) (Interstate Commerce Commission’s power to approve railroad consolidations). In short, we have “almost never felt qualified to second-guess Congress regarding the permissible degree of policy judgment that can be left to those executing or apply*475ing the law.” Mistretta v. United States, 488 U. S. 361, 416 (1989) (Scalia, J., dissenting); see id., at 373 (majority opinion).

It is true enough that the degree of agency discretion that is acceptable varies according to the scope of the power eon-gressionally conferred. See Loving v. United States, 517 U. S., at 772-773; United States v. Mazurie, 419 U. S. 544, 556-557 (1975). While Congress need not provide any direction to the EPA regarding the manner in which it is to define “country elevators,” which are to be exempt from new-stationary-source regulations governing grain elevators, see 42 U. S. C. §7411(i), it must provide substantial guidance on setting air standards that affect the entire national economy. But even in sweeping regulatory schemes we have never demanded, as the Court of Appeals did here, that statutes provide a “determinate criterion” for saying “how much [of the regulated harm] is too much.” 175 F. 3d, at 1034. In Touby, for example, we did not require the statute to decree how “imminent” was too imminent, or how “necessary” was necessary enough, or even — most relevant here — how “hazardous” was too hazardous. 500 U. S., at 165-167. Similarly, the statute at issue in Lichter authorized agencies to recoup “excess profits” paid under wartime Government contracts, yet we did not insist that Congress specify how much profit was too much. 334 U. S., at 783-786. It is therefore not conclusive for delegation purposes that, as respondents argue, ozone and particulate matter are “nonthreshold” pollutants that inflict a continuum of adverse health effects at any airborne concentration greater than zero, and hence require the EPA to make judgments of degree. “[A] certain degree of discretion, and thus of lawmaking, inheres in most executive or judicial action.” Mistretta v. United States, supra, at 417 (Scalia, J., dissenting) (emphasis deleted); see 488 U. S., at 378-379 (majority opinion). Section 109(b)(1) of the CAA, which to repeat we interpret as requiring the EPA to set air quality standards at the level that is “requi*476site” — that is, not lower or higher than is necessary — to protect the public health with an adequate margin of safety, fits comfortably within the scope of discretion permitted by our precedent.

We therefore reverse the judgment of the Court of Appeals remanding for reinterpretation that would avoid a supposed delegation of legislative power. It will remain for the Court of Appeals — on the remand that we direct for other reasons — to dispose of any other preserved challenge to the NAAQS under the judicial-review provisions contained in 42 U. S. C. § 7607(d)(9).

IV

The final two issues on which we granted certiorari concern the EPA’s authority to implement the revised ozone NAAQS in areas whose ozone levels currently exceed the maximum level permitted by that standard. The CAA designates such areas “nonattainment,” § 107(d)(1), 42 U. S. C. § 7407(d)(1); see also Pub. L. 105-178, §6103, 112 Stat. 465 (setting timeline for new ozone designations), and it exposes them to additional restrictions over and above the implementation requirements imposed generally by §110 of the CAA. These additional restrictions are found in the five substantive subparts of Part D of Title I, 42 U. S. C. §§7501-7515. Subpart 1, §§7501-7509a, contains general nonattainment regulations that pertain to every pollutant for which a NAAQS exists. Subparts 2 through 5, §§ 7511— 7514a, contain rules tailored to specific individual pollutants. Subpart 2, added by the Clean Air Act Amendments of 1990, §103, 104 Stat. 2423, addresses ozone. 42 U. S. C. §§ 7511— 75111 The dispute before us here, in a nutshell, is whether Subpart 1 alone (as the agency determined), or rather Sub-part 2 or some combination of Subparts 1 and 2, controls the implementation of the revised ozone NAAQS in non-attainment areas.

*477A

The Administrator first urges, however, that we vacate the judgment of the Court of Appeals on this issue because it lacked jurisdiction to review the EPA’s implementation policy. Section 307(b)(1) of the CAA, 42 U. S. C. § 7607(b)(1), gives the court jurisdiction over “any . . . nationally applicable regulations promulgated, or final action taken, by the Administrator,” but the EPA argues that its implementation policy was not agency “action,” was not “final” action, and is not ripe for review. We reject each of these three contentions.

At the same time the EPA proposed the revised ozone NAAQS in 1996, it also proposed an “interim implementation policy” for the NAAQS, see 61 Fed. Reg. 65752 (1996), that was to govern until the details of implementation could be put in final form through specific “rulemaking actions.” The preamble to this proposed policy declared that “the interim implementation policy . . . represent^] EPA’s preliminary views on these issues and, while it may include various statements that States must take certain actions, these statements are made pursuant to EPA’s preliminary interpretations, and thus do not bind the States and public as a matter of law.” Ibid. If the EPA had done no more, we perhaps could accept its current claim that its action was not final. However, after the agency had accepted comments on its proposed policy, and on the same day that the final ozone NAAQS was promulgated, the White House published in the Federal Register what it titled a “Memorandum for the Administrator of the Environmental Protection Agency” that prescribed implementation procedures for the EPA to follow. 62 Fed. Reg. 38421 (1997). (For purposes of our analysis we shall assume that this memorandum was not itself action by the EPA.) The EPA supplemented this memorandum with an explanation of the implementation procedures, which it published in the explanatory preamble to its final ozone *478NAAQS under the heading, “Final decision on the primary standard.” Id., at 38873. “In light of comments received regarding the interpretation proposed in the Interim Implementation Policy,” the EPA announced, it had “reconsidered that interpretation” and settled on a new one. Ibid. The provisions of “subpart 1 of part D of Title I of the Act” will immediately “apply to the implementation of the new 8-hour [ozone] standards.” Ibid.; see also id., at 38885 (new standard to be implemented “simultaneously [with the old standard] . . . under the provisions of . . . subpart 1”). Moreover, the provisions of subpart 2 “will [also] continue to apply as a matter of law for so long as an area is not attaining the [old] 1-hour standard.” Id., at 38873. Once the area reaches attainment for the old standard, however, “the provisions of subpart 2 will have been achieved and those provisions will no longer apply.” Ibid.; see also id., at 38884-38885.

We have little trouble concluding that this constitutes final agency action subject to review under §307. The bite in the phrase “final action” (which bears the same meaning in § 307(b)(1) that it does under the Administrative Procedure Act (APA), 5 U. S. C. § 704, see Harrison v. PPG Industries, Inc., 446 U. S. 578, 586 (1980)), is not in the word “action,” which is meant to cover comprehensively every manner in which an agency may exercise its power. See FTC v. Standard Oil Co. of Cal., 449 U. S. 232, 238, n. 7 (1980). It is rather in the word “final,” which requires that the action under review “mark the consummation of the agency’s de-cisionmaking process.” Bennett v. Spear, 520 U. S. 154, 177-178 (1997). Only if the “EPA has rendered its last word on the matter” in question, Harrison v. PPG Industries, Inc., supra, at 586, is its action “final” and thus reviewable. That standard is satisfied here. The EPA’s “decision-making process,” which began with the 1996 proposal and continued with the reception of public comments, concluded *479when the agency, “in light of [these comments],” and in conjunction with a corresponding directive from the White House, adopted the interpretation of Part D at issue here. Since that interpretation issued, the EPA has refused in subsequent rulemakings to reconsider it, explaining to disappointed commenters that its earlier decision was conclusive. See 63 Fed. Reg. 31014, 31018-31019 (1998). Though the agency has not dressed its decision with the conventional procedural accoutrements of finality, its own behavior thus belies the claim that its interpretation is not final.

The decision is also ripe for our review. “Ripeness ‘requires] us to evaluate both the fitness of the issues for judicial decision and the hardship to the parties of withholding court consideration.’ ” Texas v. United States, 523 U. S. 296, 300-301 (1998) (quoting Abbott Laboratories v. Gardner, 387 U. S. 136, 149 (1967)). The question before us here is purely one of statutory interpretation that would not “benefit from further factual development of the issues presented.” Ohio Forestry Assn., Inc. v. Sierra Club, 523 U. S. 726, 733 (1998). Nor will our review “inappropriately interfere with further administrative action,” ibid., since the EPA has concluded its consideration of the implementation issue. Finally, as for hardship to the parties: The respondent States must — on pain of forfeiting to the EPA control over implementation of the NAAQS — promptly undertake the lengthy and expensive task of developing state implementation plans (SIP’s) that will attain the new, more stringent standard within five years. See 42 U. S. C. §§7410, 7502. Whether or not this would suffice in an ordinary ease brought under the review provisions of the APA, see 5 U. S. C. § 704, we have characterized the special judicial-review provision of the CAA, 42 U. S. C. § 7607(b), as one of those statutes that specifically provides for “preenforcement” review, see Ohio Forestry Assn., Inc. v. Sierra Club, supra, at 737. Such statutes, we have said, permit “judicial review directly, even before the *480concrete effects normally required for APA review are felt.” Lujan v. National Wildlife Federation, 497 U. S. 871, 891 (1990). The effects at issue here surely meet that lower standard.

Beyond all this, the implementation issue was fairly included within the challenges to the final ozone rule that were properly before the Court of Appeals. Respondents argued below that the EPA could not revise the ozone standard, because to do so would trigger the use of Subpart 1, which had been supplanted (for ozone) by the specific rules of Sub-part 2. Brief for Industry Petitioners and Intervenors in No. 97-1441 (and consolidated cases) (CADC), pp. 82-84. The EPA responded that Subpart 2 did not supplant but simply supplemented Subpart 1, so that the latter section still “applies to all nonattainment areas for all NAAQS, . . . including nonattainment areas for any revised ozone standard.” Final Brief for EPA in No. 97-1441 (and consolidated cases) (CADC), pp. 67-68. The agency later reiterated that Subpart 2 “does not supplant implementation provisions for revised ozone standards. This interpretation fully harmonizes Subpart 2 with EPA’s clear authority to revise any NAAQS.” Id., at 71. In other words, the EPA was arguing that the revised standard could be issued, despite its apparent incompatibility with portions of S'ubpart 2, because it would be implemented under Subpart 1 rather than Subpart 2. The District of Columbia Circuit ultimately agreed that Subpart 2 could be harmonized with the EPA’s authority to promulgate revised NAAQS, but not because Subpart 2 is entirely inapplicable — which is one of EPA’s assignments of error. It is unreasonable to contend, as the EPA now does, that the Court of Appeals was obligated to reach the agency’s preferred result, but forbidden to assess the reasons the EPA had given for reaching that result. The implementation issue was fairly included within respondents’ challenge to the ozone rule, which all parties agree is final agency action ripe for review.

*481B

Our approach to the merits of the parties’ dispute is the familiar one of Chevron U. S. A. Inc. v. Natural Resources Defense Council, Inc., 467 U. S. 837 (1984). If the statute resolves the question whether Subpart 1 or Subpart 2 (or some combination of the two) shall apply to revised ozone NAAQS, then “that is the end of the matter.” Id., at 842-843. But if the statute is “silent or ambiguous” with respect to the issue, then we must defer to a “reasonable interpretation made by the administrator of an agency.” Id., at 844. We cannot agree with the Court of Appeals that Sub-part 2 clearly controls the implementation of revised ozone NAAQS, see 175 F. 3d, at 1048-1050, because we find the statute to some extent ambiguous. We conclude, however, that the agency’s interpretation goes beyond the limits of what is ambiguous and contradicts what in our view is quite clear. We therefore hold the implementation policy unlawful. See AT&T Corp. v. low a Utilities Bd., 525 U. S. 366, 392 (1999).

The text of Subpart 1 at first seems to point the way to a clear answer to the question, which Subpart controls? Two sections of Subpart 1, 7502(a)(1)(C) and 7502(a)(2)(D), contain switching provisions stating that if the classification of ozone nonattainment areas is “specifically provided [for] under other provisions of [Part D],” then those provisions will control instead of Subpart l’s. Thus, it is true but incomplete to note, as the Administrator does, that the substantive language of Subpart 1 is broad enough to apply to revised ozone standards. See, e. g., § 7502(a)(1)(A) (instructing the Administrator to classify nonattainment areas according to “any revised standard, including a revision of any standard in effect on November 15, 1990”); § 7502(a)(2)(A) (setting attainment deadlines). To determine whether that language does apply one must resolve the further textual issue whether some other provision, namely Subpart 2, provides for the classification of ozone nonattainment areas. If *482it does, then according to the switching provisions of Sub-part 1 it will control.

So, does Subpart 2 provide for classifying nonattainment ozone areas under the revised standard? It unquestionably does. The backbone of the subpart is Table 1, printed in § 7511(a)(1) and reproduced in the margin here,5 which defines five categories of ozone nonattainment areas and prescribes attainment deadlines for each. Section 7511(a)(1) funnels all nonattainment areas into the table for classification, declaring that “[e]ach area designated nonattainment for ozone . . . shall be classified at the time of such designation, under table 1, by operation of law.” And once an area has been classified, “the primary standard attainment date for ozone shall be as expeditiously as practicable but not later than the date provided in table 1.” The EPA argues that this text is not as clear or comprehensive as it seems, because the title of § 7511(a) reads “Classification and attainment dates for 1989 nonattainment areas,” which suggests that Subpart 2 applies only to areas that were in nonattainment in 1989, and not to areas later designated non-*483attainment under a revised ozone standard. The suggestion must be rejected, however, because § 7511(b)(1) specifically provides for the classification of areas that were in attainment in 1989 but have subsequently slipped into nonattainment. It thus makes clear that Subpart 2 is not limited solely to 1989 nonattainment areas. This eliminates the interpretive role of the title, which may only “she[d] light on some ambiguous word or phrase in the statute itself,” Carter v. United States, 530 U. S. 255, 267 (2000) (internal quotation marks omitted) (quoting Pennsylvania Dept. of Corrections v. Yeskey, 524 U. S., at 212, in turn quoting Trainmen v. Baltimore & Ohio R. Co., 331 U. S. 519, 528-529 (1947)).

It may well be, as the EPA argues — and as the concurring opinion below on denial of rehearing pointed out, see 195 F. 3d, at 11-12 — that some provisions of Subpart 2 are ill fitted to implementation of the revised standard. Using the old 1-hour averages of ozone levels, for example, as Sub-part 2 requires, see § 7511(a)(1); 44 Fed. Reg. 8202 (1979), would produce at best an inexact estimate of the new 8-hour averages, see 40 CFR §50.10, and App. I (1999). Also, to the extent that the new ozone standard is stricter than the old one, see Reply Brief for Petitioners in No. 99-1257, p. 17 (“the stricter 8-hour NAAQS”); 62 Fed. Reg. 38856, 38858 (1997) (8-hour standard of 0.09 ppm rather than 0.08 ppm would have “generally represented] the continuation of the [old] level of protection”), the classification system of Subpart 2 contains a gap, because it fails to classify areas whose ozone levels are greater than the new standard (and thus nonattaining) but less than the approximation of the old standard codified by Table 1. And finally, Subpart 2’s method for calculating attainment dates — which is simply to count forward a certain number of years from November 15, 1990 (the date the 1990 CAA Amendments took force), depending on how far out of attainment the area started — seems to make no sense for areas that are first classified under a new standard after November 15, 1990. *484If, for example, areas were classified in the year 2000, many of the deadlines would already have expired at the time of classification.

These gaps in Subpart 2's scheme prevent us from concluding that Congress clearly intended Subpart 2 to be the exclusive, permanent means of enforcing a revised ozone standard in nonattainment areas. The statute is in our view ambiguous concerning the manner in which Subpart 1 and Subpart 2 interact with regard to revised ozone standards, and we would defer to the EPA’s reasonable resolution of that ambiguity. See FDA v. Brown & Williamson Tobacco Corp., 529 U. S., at 132; INS v. Aguirre-Aguirre, 526 U. S. 415, 424 (1999). We cannot defer, however, to the interpretation the EPA has given.

Whatever effect may be accorded the gaps in Subpart 2 as implying some limited applicability of Subpart 1, they cannot be thought to render Subpart 2’s carefully designed restrictions on EPA discretion utterly nugatory once a new standard has been promulgated, as the EPA has concluded. The principal distinction between Subpart 1 and Subpart 2 is that the latter eliminates regulatory discretion that the former allowed. While Subpart 1 permits the EPA to establish classifications for nonattainment areas, Subpart 2 classifies areas as a matter of law based on a table. Compare § 7502(a)(1) with § 7511(a)(1) (Table 1). Whereas the EPA has discretion under Subpart 1 to extend attainment dates for as long as 12 years, under Subpart 2 it may grant no more than 2 years’ extension. Compare §§ 7502(a)(2)(A) and (C) with § 7511(a)(5). Whereas Subpart 1 gives the EPA considerable discretion to shape nonattainment programs, Subpart 2 prescribes large parts of them by law. Compare §§ 7502(c) and (d) with § 7511a. Yet according to the EPA, Subpart 2 was simply Congress’s “approach to the implementation of the [old] 1-hour” standard, and so there was no reason that “the new standard could not simultaneously be implemented under . . . subpart 1.” 62 Fed. Reg. *48538856, 38885 (1997); see also id., at 38873 (“[T]he provisions of subpart 1 . . . would apply to the implementation of the new 8-hour ozone standards”). To use a few apparent gaps in Subpart 2 to render its textually explicit applicability to nonattainment areas under the new standard utterly inoperative is to go over the edge of reasonable interpretation. The EPA may not construe the statute in a way that completely nullifies textually applicable provisions meant to limit its discretion.

The EPA’s interpretation making Subpart 2 abruptly obsolete is all the more astonishing because Subpart 2 was obviously written to govern implementation for some time. Some of the elements required to be included in SIP’s under Subpart 2 were not to take effect until many years after the passage of the CAA. See § 7511a(e)(3) (restrictions on “electric utility and industrial and commercial boilerfs]” to be “effective 8 years after November 15,1990”); §7511a(c)(5)(A) (vehicle monitoring program to “[b]egi[n] 6 years after November 15, 1990”); §7511a(g)(l) (emissions milestone requirements to be applied “6 years after November 15, 1990, and at intervals of every 3 years thereafter”). A plan reaching so far into the future was not enacted to be abandoned the next time the EPA reviewed the ozone standard — which Congress knew could happen at any time, since the technical staff papers had already been completed in late 1989. See 58 Fed. Reg. 13008, 13010 (1993); see also 42 U. S. C. § 7409(d)(1) (NAAQS must be reviewed and, if appropriate, revised at least once every five years). Yet nothing in the EPA’s interpretation would have prevented the agency from aborting Subpart 2 the day after it was enacted. Even now, if the EPA’s interpretation were correct, some areas of the country could be required to meet the new, more stringent ozone standard in at most the same time that Sub-part 2 had allowed them to meet the old standard. Compare § 7502(a)(2) (Subpart 1 attainment dates) with § 7511(a) (Sub-part 2 attainment dates). Los Angeles, for instance, “would *486be required to attain the revised NAAQS under Subpart 1 no later than the same year that marks the outer time limit for attaining Subpart 2’s one-hour ozone standard.” Brief for Petitioners in No. 99-1257, p. 49. An interpretation of Subpart 2 so at odds with its structure and manifest purpose cannot be sustained.

We therefore find the EPA’s implementation policy to be unlawful, though not in the precise respect determined by the Court of Appeals. After our remand, and the Court of Appeals’ final disposition of these cases, it is left to the EPA to develop a reasonable interpretation of the nonattainment implementation provisions insofar as they apply to revised ozone NAAQS.

* * *

To summarize our holdings in these unusually complex cases: (1) The EPA may not consider implementation costs in setting primary and secondary NAAQS under § 109(b) of the CAA. (2) Section 109(b)(1) does not delegate legislative power to the EPA in contravention of Art. I, § 1, of the Constitution. (3) The Court of Appeals had jurisdiction to review the EPA’s interpretation of Part D of Title I of the CAA, relating to the implementation of the revised ozone NAAQS. (4) The EPA’s interpretation of that Part is unreasonable.

The judgment of the Court of Appeals is affirmed in part and reversed in part, and the cases are remanded for proceedings consistent with this opinion.

It is so ordered.

Justice Thomas,

concurring.

I agree with the majority that §109’s directive to the agency is no less an “intelligible principle” than a host of other directives that we have approved. Ante, at 474-476. I also agree that the Court of Appeals’ remand to the agency to make its own corrective interpretation does not accord with our understanding of the delegation issue. Ante, at 472-473. I write separately, however, to express my con*487cern that there may nevertheless be a genuine constitutional problem with § 109, a problem which the parties did not address.

The parties to these cases who briefed the constitutional issue wrangled over constitutional doctrine with barely a nod to the text of the Constitution. Although this Court since 1928 has treated the “intelligible principle” requirement as the only constitutional limit on congressional grants of power to administrative agencies, see J W. Hampton, Jr., & Co. v. United States, 276 U. S. 394, 409 (1928), the Constitution does not speak of “intelligible principles.” Rather, it speaks in much simpler terms: “All legislative Powers herein granted shall be vested in a Congress.” U. S. Const., Art. 1, § 1 (emphasis added). I am not convinced that the intelligible principle doctrine serves to prevent all cessions of legislative power. I believe that there are cases in which the principle is intelligible and yet the significance of the delegated decision is simply too great for the decision to be called anything other than “legislative.”

As it is, none of the parties to these cases has examined the text of the Constitution or asked us to reconsider our precedents on cessions of legislative power. On a future day, however, I would be willing to address the question whether our delegation jurisprudence has strayed too far from our Founders’ understanding of separation of powers.

Justice Stevens,

with whom Justice Souter joins, concurring in part and concurring in the judgment.

Section 109(b)(1) delegates to the Administrator of the Environmental Protection Agency (EPA) the authority to promulgate national ambient air quality standards (NAAQS). In Part III of its opinion, ante, at 472-476, the Court convincingly explains why the Court of Appeals erred when it concluded that § 109 effected “an unconstitutional delegation of legislative power.” American Trucking Assns., Inc. v. EPA, 175 F. 3d 1027, 1033 (CADC 1999) (per curiam). *488I wholeheartedly endorse the Court’s result and endorse its explanation of its reasons, albeit with the following caveat.

The Court has two choices. We could choose to articulate our ultimate disposition of this issue by frankly acknowledging that the power delegated to the EPA is “legislative” but nevertheless conclude that the delegation is constitutional because adequately limited by the terms of the authorizing statute. Alternatively, we could pretend, as the Court does, that the authority delegated to the EPA is somehow not “legislative power.” Despite the fact that there is language in our opinions that supports the Court’s articulation of our holding,1 I am persuaded that it would be both wiser and more faithful to what we have actually done in delegation cases to admit that agency rulemaking authority is “legislative power.”2

The proper characterization of governmental power should generally depend on the nature of the power, not on the identity of the person exercising it. See Black’s Law Dictionary 899 (6th ed. 1990) (defining “legislation” as, inter alia, “[formulation of rule[s] for the future”); 1 K. Davis & R. Pierce, Administrative Law Treatise §2.3, p. 37 (3d ed. 1994) (“If legislative power means the power to make rules of conduct that bind everyone based on resolution of major policy issues, scores of agencies exercise legislative power routinely by *489promulgating what are candidly called ‘legislative rules’”). If the NAAQS that the EPA promulgated had been prescribed by Congress, everyone would agree that those rules would be the product of an exercise of “legislative power.” The same characterization is appropriate when an agency exercises rulemaking authority pursuant to a permissible delegation from Congress.

My view is not only more faithful to normal English usage, but is also fully consistent with the text of the Constitution. In Article I, the Framers vested “All legislative Powers” in the Congress, Art. I, § 1, just as in Article II they vested the “executive Power” in the President, Art. II, § 1. Those provisions do not purport to limit the authority of either recipient of power to delegate authority to others. See Bowsher v. Synar, 478 U. S. 714, 752 (1986) (Stevens, J., concurring in judgment) (“Despite the statement in Article I of the Constitution that ‘All legislative powers herein granted shall be vested in a Congress of the United States,’ it is far from novel to acknowledge that independent agencies do indeed exercise legislative powers”); INS v. Chadha, 462 U. S. 919, 985-986 (1983) (White, J., dissenting) (“[Ljegisla-tive power can be exercised by independent agencies and Executive departments ...”); 1 Davis & Pierce, Administrative Law Treatise § 2.6, at 66 (“The Court was probably mistaken from the outset in interpreting Article I’s grant of power to Congress as an implicit limit on Congress’ authority to delegate legislative power”). Surely the authority granted to members of the Cabinet and federal law enforcement agents is properly characterized as “Executive” even though not exercised by the President. Cf. Morrison v. Olson, 487 U. S. 654, 705-706 (1988) (Scalia, J., dissenting) (arguing that the independent counsel exercised “executive power” unconstrained by the President).

It seems clear that an executive agency’s exercise of rule-making authority pursuant to a valid delegation from Congress is “legislative.” As long as the delegation provides a *490sufficiently intelligible principle, there is nothing inherently unconstitutional about it. Accordingly, while I join Parts I, II, and IV of the Court’s opinion, and agree with almost everything said in Part III, I would hold that when Congress enacted § 109, it effected a constitutional delegation of legislative power to the EPA.

Justice Breyer,

concurring in part and concurring in the judgment.

I join Parts I, III, and IV of the Court’s opinion. I also agree with the Court’s determination in Part II that the Clean Air Act does not permit the Environmental Pi’otection Agency to consider the economic costs of implementation when setting national ambient air quality standards under § 109(b)(1) of the Act. But I would not rest this conclusion solely upon § 109’s language or upon a presumption, such as the Court’s presumption that any authority the Act grants the EPA to consider costs must flow from a “textual commitment” that is “clear.” Ante, at 468. In order better to achieve regulatory goals — for example, to allocate resources so that they save more lives or produce a cleaner environment — regulators must often take account of all of a proposed regulation’s adverse effects, at least where those adverse effects clearly threaten serious and disproportionate public harm. Hence, I believe that, other things being equal, we should read silences or ambiguities in the language of regulatory statutes as permitting, not forbidding, this type of rational regulation.

In these cases, however, other things are not equal. Here, legislative history, along with the statute’s structure, indicates that § 109’s language reflects a congressional decision not to delegate to the agency the legal authority to consider economic costs of compliance.

For one thing, the legislative history shows that Congress intended the statute to be “technology forcing.” Senator Edmund Muskie, the primary sponsor of the 1970 amend*491ments to the Act, introduced them by saying that Congress’ primary responsibility in drafting the Act was not “to be limited by what is or appears to be technologically or economically feasible,” but “to establish what the public interest requires to protect the health of persons,” even if that means that “industries will be asked to do what seems to be impossible at the present time” 116 Cong. Rec. 32901-32902 (1970), 1 Legislative History of the Clean Air Amendments of 1970 (Committee Print compiled for the Senate Committee on Public Works by the Library of Congress), Ser. No. 93-18, p. 227 (1974) (hereinafter Leg. Hist.) (emphasis added).

The Senate directly focused upon the technical feasibility and cost of implementing the Act’s mandates. And it made clear that it intended the Administrator to develop air quality standards set independently of either. The Senate Report for the 1970 amendments explains:

“In the Committee discussions, considerable concern was expressed regarding the use of the concept of technical feasibility as the basis of ambient air standards. The Committee determined that 1) the health of people is more important than the question of whether the early achievement of ambient air quality standards protective of health is technically feasible; and, 2) the growth of pollution load in many areas, even with application of available technology, would still be deleterious to public health. ...

“Therefore, the Committee determined that existing sources of pollutants either should meet the standard of the law or be closed down_” S. Rep. No. 91-1196, pp. 2-3 (1970), 1 Leg. Hist. 402-403 (emphasis added).

Indeed, this Court, after reviewing the entire legislative history, concluded that the 1970 amendments were “expressly designed to force regulated sources to develop pollution control devices that-might at the time appear to be economically or technologically infeasible.” Union Elec. Co. *492v. EPA, 427 U. S. 246, 257 (1976) (emphasis added). And the Court added that the 1970 amendments were intended to be a “drastic remedy to ... a serious and otherwise uncheckable problem.” Id., at 256. Subsequent legislative history confirms that the technology-forcing goals of the 1970 amendments are still paramount in today’s Act. See Clean Air Conference Report (1977): Statement of Intent; Clarification of Select Provisions, 123 Cong. Rec. 27070 (1977) (stating, regarding the 1977 amendments to the Act, that “this year’s legislation retains and even strengthens the technology forcing . . . goals of the 1970 Act”); S. Rep. No. 101-228, p. 5 (1989) (stating that the 1990 amendments to the Act require ambient air quality standards to be set at “the level that ‘protects the public health’ with an ‘adequate margin of safety,’ without regard to the economic or technical feasibility of attainment” (emphasis added)).

To read this legislative history as meaning what it says does not impute to Congress an irrational intent. Technology-forcing hopes can prove realistic. Those persons, for example, who opposed the. 1970 Act’s insistence on a 90% reduction in auto emission pollutants, on the ground of excessive cost, saw the development of catalytic converter technology that helped achieve substantial reductions without the economic catastrophe that some had feared. See §6(a) of the Clean Air Act Amendments of 1970, amending §§ 202(b)(1)(A), (B), 84 Stat. 1690 (codified at 42 U.S.C. §§ 7521(b)(1)(A), (B)) (requiring a 90% reduction in emissions); 1 Leg. Hist. 238, 240 (statement of Sen. Griffin) (arguing that the emissions standards could “force [the automobile] industry out of existence” because costs “would not be taken into account”); see generally Reitze, Mobile Source Air Pollution Control, 6 Env. Law. 309,326-327 (2000) (discussing the development of the catalytic converter).

At the same time, the statute’s technology-forcing objective makes regulatory efforts to determine the costs of implementation both less important and more difficult. It *493means that the relevant economic costs are speculative, for they include the cost of unknown future technologies. It also means that efforts to take costs into account can breed time-consuming and potentially unresolvable arguments about the accuracy and significance of cost estimates. Congress could have thought such efforts not worth the delays and uncertainties that would accompany them. In any event, that is what the statute’s history seems to say. See Union Elec., supra, at 256-259. And the matter is one for Congress to decide.

Moreover, the Act does not, on this reading, wholly ignore cost and feasibility. As the majority points out, ante, at 466-467, the Act allows regulators to take those concerns into account when they determine how to implement ambient air quality standards. Thus, States may consider economic costs when they select the particular control devices used to meet the standards, and industries experiencing difficulty in reducing their emissions can seek an exemption or variance from the state implementation plan. See Union Elec., supra, at 266 (“[T]he most important forum for consideration of claims of economic and technological infeasibility is before the state agency formulating the implementation plan”).

The Act also permits the EPA, within certain limits, to consider costs when it sets deadlines by which areas must attain the ambient air quality standards. 42 U. S. C. § 7502(a)(2)(A) (providing that “the Administrator may extend the attainment date ... for a period no greater than 10 years from the date of designation as nonattainment, considering the severity of nonattainment and the availability and feasibility of pollution control measures”); § 7502(a)(2)(C) (permitting the Administrator to grant up to two additional 1-year extensions); cf. §§ 7511(a)(1), (5) (setting more rigid attainment deadlines for areas in nonattainment of the ozone standard, but permitting the Administrator to grant up to two 1-year extensions). And Congress can change those statutory limits if necessary. Given the ambient air quality *494standards' substantial effects on States, cities, industries, and their suppliers and customers, Congress will hear from those whom compliance deadlines affect adversely, and Congress can consider whether legislative change is warranted. See, e. g., Steel Industry Compliance Extension Act of 1981, 95 Stat. 139 (codified at 42 U. S. C. § 7413(e) (1988 ed.)) (repealed 1990) (granting the Administrator discretion to extend the ambient air quality standard attainment date set in the 1977 Act by up to three years for steelmaking facilities).

Finally, contrary to the suggestion of the Court of Appeals and of some parties, this interpretation of §109 does not require the EPA to eliminate every health risk, however slight, at any economic cost, however great, to the point of “hurtling” industry over “the brink of ruin,” or even forcing “deindustrialization.” American Trucking Assns., Inc. v. EPA, 175 F. 3d 1027, 1037, 1038, n. 4 (CADC 1999); see also Brief for Cross-Petitioners in No. 99-1426, p. 25. The statute, by its express terms, does not compel the elimination of all risk; and it grants the Administrator sufficient flexibility to avoid setting ambient air quality standards ruinous to industry.

Section 109(b)(1) directs the Administrator to set standards that are “requisite to protect the public health” with “an adequate margin of safety.” But these words do not describe a world that is free of all risk — an impossible and undesirable objective. See Industrial Union Dept., AFL-CIO v. American Petroleum Institute, 448 U. S. 607, 642 (1980) (plurality opinion) (the word “safe” does not mean “risk-free”). Nor are the words “requisite” and “public health” to be understood independent of context. We consider football equipment “safe” even if its use entails a level of risk that would make drinking water “unsafe” for consumption. And what counts as “requisite” to protecting the public health will similarly vary with background circumstances, such as the public’s ordinary tolerance of the particular health risk in the particular context at issue. The Administrator can *495consider such background circumstances when “deciding] what risks are acceptable in the world in which we live.” Natural Resources Defense Council, Inc. v. EPA, 824 F. 2d 1146, 1165 (CADC 1987).

The statute also permits the Administrator to take account of comparative health risks. That is to say, she may consider whether a proposed rule promotes safety overall. A rule likely to cause more harm to health than it prevents is not a rule that is “requisite to protect the public health.” For example, as the Court of Appeals held and the parties do not contest, the Administrator has the authority to determine to what extent possible health risks stemming from reductions in tropospheric ozone (which, it is claimed, helps prevent cataracts and skin cancer) should be taken into account in setting the ambient air quality standard for ozone. See 175 F. 3d, at 1050-1053 (remanding for the Administrator to make that determination).

The statute ultimately specifies that the standard set must be “requisite to protect the public health” “in the judgment of the Administrator,” § 109(b)(1), 84 Stat. 1680 (emphasis added), a phrase that grants the Administrator considerable discretionary standard-setting authority.

The statute’s words, then, authorize the Administrator to consider the severity of a pollutant’s potential adverse health effects, the number of those likely to be affected, the distribution of the adverse effects, and the uncertainties surrounding each estimate. Cf. Sunstein, Is the Clean Air Act Unconstitutional?, 98 Mich. L. Rev. 303, 364 (1999). They permit the Administrator to take account of comparative health consequences. They allow her to take account of context when determining the acceptability of small risks to health. And they give her considerable discretion when she does so.

This discretion would seem sufficient to avoid the extreme results that some of the industry parties fear. After all, the EPA, in setting standards that “protect the public health” *496with “an adequate margin of safety,” retains discretionary authority to avoid regulating risks that it reasonably concludes are trivial in context. . Nor need regulation lead to deindustrialization. Preindustrial society was not a very healthy society; hence a standard demanding the return of the Stone Age would not prove “requisite to protect the public health.”

Although I rely more heavily than does the Court upon legislative history and alternative sources of statutory flexibility, I reach the same ultimate conclusion. Section 109 does not delegate to the EPA authority to base the national ambient air quality standards, in whole or in part, upon the economic costs of compliance.

*249Mr. Justice Marshall

delivered the opinion of the Court.

After the Administrator of the Environmental Protection Agency (EPA) approves a state implementation plan under the Clean Air Act, the plan may be challenged in a court of appeals within 30 days, or after 30 days have run if newly discovered or available information justifies subsequent review. We must decide whether the operator of a regulated emission source, in a petition for review of an EPA-approved state plan filed after the original 30-day appeal period, can raise the claim that it is economically or technologically infeasible to comply with the plan.

I

We have addressed the history and provisions of the Clean Air Amendments of 1970, Pub. L. 91-604, 84 Stat. 1676, in detail in Train v. Natural Resources Defense Council (NRDC), 421 U. S. 60 (1975), and will not repeat that discussion here. Suffice it to say that the Amendments reflect congressional dissatisfaction with the progress of existing air pollution programs and a determination to “tak[e] a stick to the States,” id., at 64, in order to guarantee the prompt attainment and maintenance of specified air quality standards. The heart of the Amendments is the requirement that each State formulate, subject to EPA approval, an implementation plan designed to achieve national primary ambient air quality standards — those necessary to protect the public health — “as expeditiously as practicable but . . . in no case later than three years from the date of ap*250proval of such plan.” § 110 (a) (2) (A) of the Clean Air Act, as added, 84 Stat. 1680, 42 U. S. C. § 1857&-5 (a) (2) (A). The plan must also provide for the attainment of national secondary ambient air quality standards'— those necessary to protect the public welfare — within a “reasonable time.” Ibid. Each State is given wide discretion in formulating its plan, and the Act provides that the Administrator “shall approve” the proposed plan if it has been adopted after public notice and hearing and if it meets eight specified criteria. § 110 (a) (2).1

*251On April 30, 1971, the Administrator promulgated national primary and secondary standards for six air pollutants he found to have an adverse effect on the public health and welfare. 40 CFR pt. 50 (1975). See § 108 (a) of the Act, as added, 84 Stat. 1678, 42 U. S. C. § 1857c-3 (a). Included among them was sulfur dioxide, at issue here. 40 CFR §§ 50.4-50.5 (1975). After the promulgation of the national standards, the State of Missouri formulated its implementation plan and submitted it for approval. Since sulfur dioxide levels exceeded national primary standards in only one of the *252State’s five air quality regions — the Metropolitan St. Louis Interstate region, 40 CFR § 52.1321 (1975)- — the Missouri plan concentrated on a control strategy and regulations to lower emissions in that area. The plan’s emission limitations were effective at once, but the State retained authority to grant variances to particular sources that could not immediately comply.2 Mo. Rev. Stat. §203.110 (1972). The Administrator approved the plan on May 31, 1972. See 40 CFR § 52.1320 et seq. (1975).

Petitioner is an electric utility company servicing the St. Louis metropolitan area, large portions of Missouri, and parts of Illinois and Iowa. Its three coal-fired generating plants in the metropolitan St. Louis area are subject to the sulfur dioxide restrictions in the Missouri implementation plan. Petitioner did not seek review of the Administrator’s approval of the plan within 30 days, as it was entitled to do under § 307 (b)(1) of the Act, as added, 84 Stat. 1708, 42 U. S. C. § 1857h-5 (b) (1), but rather applied to the appropriate state and county agencies for variances from the emission limitations affecting its three plants. Petitioner received one-year variances, which could be extended upon reapplication. The variances on two of petitioner’s three plants had expired and petitioner was applying for extensions when, on May 31, 1974, the Administrator notified petitioner that sulfur dioxide emissions from its plants violated the emission limitations contained in the Missouri plan.3 See 40 Fed. Reg. 3566 (1975). Shortly there*253after petitioner filed a petition in the Court of Appeals for the Eighth Circuit for review of the Administrator’s 1972 approval of the Missouri implementation plan.

Section 307 (b)(1) allows petitions for review to be filed in an appropriate court of appeals more than 30 days after the Administrator’s approval of an implementation plan only if the petition is "based solely on grounds arising- after such 30th day.” Petitioner claimed to meet this requirement by asserting, inter alia, that various economic and technological difficulties had arisen more than 30 days after the Administrator’s approval and that these difficulties made compliance with the emission limitations impossible.4 The Court of Ap*254peals ordered briefing on the question of its subject-matter jurisdiction to hear the case and, after argument, granted the motions of the EPA and intervenor-respond-ents, the Attorney General of Missouri and the Missouri Air Conservation Commission, to dismiss the petition for review for lack of jurisdiction.

The court held that “only matters which, if known to the Administrator at the time of his action [in approving a state implementation plan], would justify setting aside that action are properly re viewable after the initial 30 day review period.” 515 F. 2d 206, 216 (1975). Since, in the court’s view, claims of economic and technological infeasibility could not properly provide a basis for the Administrator’s rejecting a plan, such claims could not serve — at any time — as the basis for a court’s overturning an approved plan. Accordingly, insofar as petitioner’s claim of newly discovered or available information was grounded on an assertion of economic and technological infeasibility, the court held itself to be without jurisdiction to consider the petition for review, and so dismissed the petition. In so holding the Court of Appeals considered and rejected the contrary or partially contrary holdings of three other Circuits. Buckeye Power, Inc. v. EPA, 481 F. 2d 162, 168-169 (CA6 1973) (but see id., at 173); Appalachian Power Co. v. EPA, 477 F. 2d 495, 505-507 (CA4 1973); Duquesne Light Co. v. EPA, 481 F. 2d 1 (CA3 1973); Getty Oil Co. v. Ruckelshaus, 467 F. 2d 349 (CA3 1972), cert. denied, 409 U. S. 1125 (1973). See also St. Joe Minerals Corp. v. EPA, 508 F. 2d 743, 746-749 (CA3 1975), vacated and remanded, *255425 U. S. 987 (1976); Duquesne Light Co. v. EPA, 522 F. 2d 1186 (CA3 1975), cert. pending, No. 75-736. On the other hand, the Eighth Circuit found support for its position in the decisions of several other Circuits. South Terminal Corp. v. EPA, 504 F. 2d 646, 675-676 (CA1 1974); Texas v. EPA, 499 F. 2d 289, 317 (CA5 1974); Natural Resources Defense Council v. EPA, 507 F. 2d 905, 914 (CA9 1974). See also Indiana & Michigan Electric Co. v. EPA, 509 F. 2d 839, 843-845 (CA7 1975). Cf. Buckeye Power, Inc. v. EPA, 525 F. 2d 80 (CA6 1975). We granted certiorari to resolve the conflict among the Circuits, 423 U. S. 821 (1975), and we now affirm.

II

A

We reject at the outset petitioner’s suggestion that a claim of economic or technological infeasibility may be considered upon a petition for review based on new information and filed more than 30 days after approval of an implementation plan even if such a claim could not be considered by the Administrator in approving a plan or by a court in reviewing a plan challenged within the original 30-day appeal period. In pertinent part § 307 (b)(1) provides:

“A petition for review of the Administrator’s action in approving or promulgating any implementation plan under section 110 .. . may be filed only in the United States Court of Appeals for the appropriate circuit. Any such petition shall be filed within 30 days from the date of such promulgation or approval, or after such date if such petition is based solely on grounds arising after such 30th day.”

Regardless of when a petition for review is filed under i 307 (b)(1), the court is limited to reviewing “the Ad*256ministrator’s action in approving . . . [the] implementation plan . . . Accordingly, if new “grounds” are alleged, they must be such that, had they been known at the time the plan was presented to the Administrator for approval, it would have been an abuse of discretion for the Administrator to approve the plan. To hold otherwise would be to transfer a substantial responsibility in administering the Clean Air Act from the Administrator and the state agencies to the federal courts.

B

Since a reviewing court — regardless of when the petition for review is filed — may consider claims of economic and technological infeasibility only if the Administrator may consider such claims in approving or rejecting a state implementation plan, we must address ourselves to the scope of the Administrator’s responsibility. The Administrator’s position is that he has no power whatsoever to reject a state implementation plan on the ground that it is economically or technologically infeasible, and we have previously accorded great deference to the Administrator’s construction of the Clean Air Act. See Train v. NRDC, 421 U. S., at 75. After surveying the relevant provisions of the Clean Air Amendments of 1970 and their legislative history, we agree that Congress intended claims of economic and technological infeasibility to be wholly foreign to the Administrator’s consideration of a state implementation plan.

As we have previously recognized, the 1970 Amendments to the Clean Air Act were a drastic remedy to what was perceived as a serious and otherwise uncheckable problem of air pollution. The Amendments place the primary responsibility for formulating pollution control strategies on the States, but nonetheless subject *257the States to strict minimum compliance requirements. These requirements are of a "technology-forcing character,” Train v. NRDC, supra, at 91, and are expressly designed to force regulated sources to develop pollution control devices that might at the time appear to be economically or technologically infeasible.

This approach is apparent on the face of § 110 (a)(2). The provision sets out eight criteria that an implementation plan must satisfy, and provides that if these criteria are met and if the plan was adopted after reasonable notice and hearing, the Administrator "shall approve” the proposed state plan. The mandatory "shall” makes it quite clear that the Administrator is not to be concerned with factors other than those specified, Train v. NRDC, supra, at 71 n. 11, 79, and none of the eight factors appears to permit consideration of technological or economic infeasibility.5 Nonetheless, if a basis is to be found for allowing the Administrator to consider such claims, it must be among the eight criteria, and so it is here that the argument is focused.

It is suggested that consideration of claims of technological and economic infeasibility is required by the first criterion — that the primary air quality standards be met “as expeditiously as practicable but... in no case later than three years . . .” and that the secondary air *258quality standards be met within a “reasonable time.” §110 (a)(2)(A). The argument is that what is “practicable” or “reasonable” cannot be determined without assessing whether what is proposed is possible. This argument does not survive analysis.

Section 110 (a)(2)(A)’s three-year deadline for achieving primary air quality standards is central to the Amendments’ regulatory scheme and, as both the language and the legislative history of the requirement make clear, it leaves no room for claims of technological or economic infeasibility. The 1970 congressional debate on the Amendments centered on whether technology forcing was necessary and desirable in framing and attaining air quality standards sufficient to protect the public health, standards later termed primary standards. The House version of the Amendments was quite moderate in approach, requiring only that health-related standards be met “within a reasonable time.” H. R. 17255, 91st Cong., 2d Sess., § 108 (c)(1)(C)(i) (1970). The Senate bill, on the other hand, flatly required that, possible or not, health-related standards be met “within three years.” S. 4358, 91st Cong., 2d Sess., § 111 (a)(2) (A) (1970).

The Senate’s stiff requirement was intended to foreclose the claims of emission sources that it would be economically or technologically infeasible for them to achieve emission limitations sufficient to protect the public health within the specified time. As Senator Muskie, manager of the Senate bill, explained to his chamber:

“ ‘The first responsibility of Congress is not the making of technological or economic judgments— or even to be limited by what is or appears to be technologically or economically feasible. Our responsibility is to establish what the public interest requires to protect the health of persons. This may *259mean that people and industries will be asked to do what seems to be impossible at the present time.5 ” 116 Cong. Rec. 32901-32902 (1970).

See also id., at 32919 (remarks of Sen. Cooper); 33115 (remarks of Sen. Prouty). This position reflected that of the Senate committee:

“In the Committee discussions, considerable concern was expressed regarding the use of the concept of technical feasibility as the basis of ambient air standards. The Committee determined that 1) the health of people is more important than the question of whether the early achievement of ambient air quality standards protective of health is technically feasible; and 2) the growth of pollution load in many areas, even with application of available technology, would still be deleterious to public health.

“Therefore, the Committee determined that existing sources of pollutants either should meet the standard of the law or be closed down ....” S. Rep. No. 91-1196, pp. 2-3 (1970).

The Conference Committee and, ultimately, the entire Congress accepted the Senate’s three-year mandate for the achievement of primary air quality standards, and the clear import of that decision is that the Administrator must approve a plan that provides for attainment of the primary standards in three years even if attainment does not appear feasible. In rejecting the House’s version of reasonableness, however, the conferees strengthened the Senate version. The Conference Committee made clear that the States could not procrastinate until the deadline approached. Rather, the primary standards had to be met in less than three years if possible; they had to be met “as expeditiously as practica*260ble.” § 110 (a)(2)(A). Whatever room there is for considering claims of infeasibility in the attainment of primary standards must lie in this phrase, which is, of course, relevant only in evaluating those implementation plans that attempt to achieve the primary standard in less than three years.

It is argued that when such a state plan calls for proceeding more rapidly than economics and the available technology appear to allow, the plan must be rejected as not “practicable.” Whether this is a correct reading of § 110 (a) (2) (A) depends on how that section’s “as expeditiously as practicable” phrase is characterized. The Administrator’s position is that § 110 (a) (2) (A) sets only a minimum standard that the States may exceed in their discretion, so that he has no power to reject an infeasible state plan that surpasses the minimum federal requirements — a plan that reflects a state decision to engage in technology forcing on its own and to proceed more expeditiously than is practicable. On the other hand, petitioner and amici supporting its position argue that § 110 (a) (2) (A) sets a mandatory standard that the States must meet precisely, and conclude that the Administrator may reject a plan for being too strict as well as for being too lax. Since the arguments supporting this theory are also made to show that the Administrator must reject a state plan that provides for achieving more than the secondary air quality standards require, we defer consideration of this question in order to outline the development and content of the secondary standards provision of § 110 (a)(2)(A).

Secondary air quality standards, those necessary to protect the public welfare, were subject to far less legislative debate than the primary standards. The House version of the Amendments treated welfare-related standards together with health-related standards, and *261required both to be met "within a reasonable time.” H. R. 17255, 91st Cong., 2d Sess., §§ 107 (e)(1), 108 (c) (l)(C)(i) (1970). The Senate bill, on the other hand, treated health- and welfare-related standards separately and did not require that welfare-related standards be met in any particular time at all, S. 4358, 91st Cong., 2d Sess., §§ 110 (a)(3), 110 (b), 111 (a)(2)(A) (1970), although the Committee Report expressed the desire that they be met “as rapidly as possible.” S. Rep. No. 91-1196, p. 11 (1970). The final Amendments also separated welfare-related standards from health-related standards, labeled them secondary air quality standards, and adopted the House’s requirement that they be met within a “reasonable time.” §§ 109 (b), 110 (a)(2)(A). Thus, technology forcing is not expressly required in achieving standards to protect the public welfare.6

It does not necessarily follow, however, that the Administrator may consider claims of impossibility in assessing a state plan for achieving secondary standards. As with plans designed to achieve primary standards in less than three years, the scope of the Administrator’s power to reject a plan depends on whether the State itself may decide to engage in technology forcing and adopt a plan more stringent than federal law demands.7

Amici Appalachian Power Co. et al. argue that the Amendments do not give such broad power to the States. *262They claim that the States are precluded from submitting implementation plans more stringent than federal law demands by § 110 (a) (2)’s second criterion — that the plan contain such control devices “as may be necessary” to achieve the primary and secondary air quality standards. §110(a)(2)(B).8 The contention is that an overly restrictive plan is not “necessary” for attainment of the national standards and so must be rejected by the Administrator.9

The principal support for this theory of amici lies in the fact that while the House and Senate versions of §110 (a)(2) both expressly provided that the States could submit for the Administrator’s approval plans that were stricter than the national standards required, see H. R. 17255, 91st Cong., 2d Sess., § 108 (c) (1970); S. 4358, 91st Cong., 2d Sess., § 111 (a)(1) (1970), the section as enacted contains no such express language. Amici argue that the Conference Committee must have decided to require state implementation plans simply — and precisely — to meet the national standards. The argument of amici proves too much. A Conference Committee lacks power to make substantive changes on matters about *263which both Houses agree. 2 U. S. C. § 190c (a) (Senate Conference Reports); Rule XXVIII (3), Rules of the House of Representatives, and § 546, Jefferson’s Manual, H. R. Doc. No. 384, 92d Cong., 2d Sess., 526, 270-271 (1972); National Coal Operators’ Assn. v. Kleppe, 423 U. S. 388, 401 n. 10 (1976). Here the Conference Report expressly notes that both the Senate and House bills would allow States to submit plans more stringent than the national standards demand, and offers no suggestion that the Conference bill intended to change that result, even if it could. H. R. Conf. Rep. No. 91-1783, p. 45 (1970). And while the final language of § 110 (a) (2)(B) may be less explicit than the versions originally approved by the House and the Senate, the most natural reading of the “as may be necessary” phrase in context is simply that the Administrator must assure that the minimal, or “necessary,” requirements are met, not that he detect and reject any state plan more demanding than federal law requires.10

This reading is further supported by practical considerations. Section 116 of the Clean Air Act, as added, 84 Stat. 1689, 42 U. S. C. § 1857d-l (1970 ed., Supp. IV), provides that the States may adopt emission standards *264stricter than the national standards.11 Amici argue that such standards must be adopted and enforced independently of the EPA-approved state implementation plan. This construction of §§ 110 and 116, however, would not only require the Administrator to expend considerable time and energy determining whether a state plan was precisely tailored to meet the federal standards,12 but would simultaneously require States desiring stricter standards to enact and enforce two sets of emission standards, one federally approved plan and one stricter state plan. We find no basis in the Amendments for visiting such wasteful burdens upon the States and the Administrator, and so we reject the argument of amici.

We read the “as may be necessary” requirement of § 110 (a) (2) (B) to demand only that the implementation plan submitted by the State meet the “minimum conditions” of the Amendments.13 Train v. NRDC, 421 *265U. S., at 71 n. 11. Beyond that, if a State makes the legislative determination that it desires a particular air quality by a certain date and that it is willing to force technology to attain it — or lose a certain industry if attainment is not possible — such a determination is fully consistent with the structure and purpose of the Amendments, and § 110 (a) (2) (B) provides no basis for the EPA Administrator to object to the determination on the ground of infeasibility.14 See Train v. NRDC, supra, at 79.

In sum, we have concluded that claims of economic or technological infeasibility may not be considered by the Administrator in evaluating a state requirement that primary ambient air quality standards be met in the mandatory three years. And, since we further conclude that the States may submit implementation plans more stringent than federal law requires and that the Administrator must approve such plans if they meet the minimum requirements of § 110 (a)(2), it follows that the language of §110 (a)(2)(B) provides no basis for the Administrator ever to reject a state implementation plan on the ground that it is economically or technologically infeasible. Accordingly, a court of appeals reviewing an *266approved plan under § 307 (b) (1) cannot set it aside on those grounds, no matter when they are raised.

Ill

Our conclusion is bolstered by recognition that the Amendments do allow claims of technological and economic infeasibility to be raised in situations where consideration of such claims will not substantially interfere with the primary congressional purpose of prompt attainment of the national air quality standards. Thus, we do not hold that claims of infeasibility are never of relevance in the formulation of an implementation plan or that sources unable to comply with emission limitations must inevitably be shut down.

Perhaps, the most important forum for consideration of claims of economic and technological infeasibility is before the state agency formulating the implementation plan. So long as the national standards are met, the State may select whatever mix of control devices it desires, Train v. NRDC, supra, at 79, and industries with particular economic or technological problems may seek special treatment in the plan itself. Cf. 40 CFR §§51.2 (b), (d) (1975); S. Rep. No. 91-1196, p. 36 (1970). Moreover, if the industry is not exempted from, or accommodated by, the original plan, it may obtain a variance, as petitioner did in this case; and the variance, if granted after notice and a hearing, may be submitted to the EPA as a revision of the plan.15 § 110 (a) (3) (A), as amended, 88 Stat. 256, 42 U. S. C. § 1857c-5 (a) (3) (A) (1970 ed., Supp. IY). Lastly, an industry denied an exemption from the implementation plan, or denied a subsequent variance, may be able to take its claims of *267economic or technological infeasibility to the state courts. See, e. g., Mo. Rev. Stat. §203.130 (1972); Cal. Health & Safety Code § 39506 (West 1973); Pa. Stat. Ann., Tit. 71, § 1710.41 (1962).16

While the State has virtually absolute power in allocating emission limitations so long as the national standards are met, if the state plan cannot meet the national standards, the EPA is implicated in any postponement procedure. There are two ways that a State can secure relief from the EPA for individual emission sources, or classes of sources, that cannot meet the national standards. First, if the Governor of the State so requests at the time the original implementation plan is submitted, and if the State provides reasonable interim controls, the Administrator may allow a two-year extension of the three-year deadline for attainment of primary air quality standards if he finds, inter alia, that it is technologically infeasible for the source to comply. § 110 (e).17 Second, again upon application of the Governor of the State, the Administrator may allow a one-year postponement of any compliance date in an implementation plan if he finds, inter alia, that compliance is technologically *268infeasible and that “the continued operation of [the emission source] is essential to national security or to the public health or welfare . . . § 110 (f). See Train v. NRDC, 421 U. S., at 81.

Even if the State does not intervene on behalf of an emission source, technological and economic factors may be considered in at least one other circumstance. When a source is found to be in violation of the state implementation plan, the Administrator may, after a conference with the operator, issue a compliance order rather than seek civil or criminal enforcement. Such an order must specify a “reasonable” time for compliance with the relevant standard, taking into account the seriousness of the violation and “any good faith efforts to comply with applicable requirements.” § 113 (a)(4) of the Clean Air Act, as added, 84 Stat. 1686, 42 U. S. C. § 1857o-8 (a) (4). Claims of technological or economic infeasibility, the Administrator agrees, are relevant to fashioning an appropriate compliance order under § 113 (a) (4). Brief for Respondent EPA 36 n. 34.18

In short, the Amendments offer ample opportunity for consideration of claims of technological and economic infeasibility. Always, however, care is taken that consideration of such claims will not interfere substantially with the primary goal of prompt attainment of the national standards. Allowing such claims to be raised by *269appealing the Administrator’s approval of an implementation plan, as petitioner suggests, would frustrate congressional intent. It would permit a proposed plan to be struck down as infeasible before it is given a chance to work, even though Congress clearly contemplated that some plans would be infeasible when proposed. And it would permit the Administrator or a federal court to reject a State’s legislative choices in regulating air pollution, even though Congress plainly left with the States, so long as the national standards were met, the power to determine which sources would be burdened by regulation and to what extent. Technology forcing is a concept somewhat new to our national experience and it necessarily entails certain risks. But Congress considered those risks in passing the 1970 Amendments and decided that the dangers posed by uncontrolled air pollution made them worth taking. Petitioner’s theory would render that considered legislative judgment a nullity, and that is a result we refuse to reach.19

Affirmed.

Mr. Justice Powell,

with whom The Chief Justice joins, concurring.

I join the opinion of the Court because the statutory scheme and the legislative history, thoroughly described in the Court’s opinion, demonstrate irrefutably that Congress did not intend to permit the Administrator of *270the Environmental Protection Agency to reject a proposed state implementation plan on the grounds of economic or technological infeasibility. Congress adopted this position despite its apparent awareness that in some cases existing sources that cannot meet the standard of the law must be closed down.1

The desire to impose strong incentives on industry to encourage the rapid development and adoption of pollution control devices is understandable. But it is difficult to believe that Congress would adhere to its absolute position if faced with the potentially devastating consequences to the public that this case vividly demonstrates.

Petitioner is an electric utility supplying power demands in the St. Louis metropolitan area, a large part of Missouri, and parts of Illinois and Iowa. It alleges that it cannot continue to operate if forced to comply with the sulfur dioxide restrictions contained in the *271Missouri implementation plan approved by the Administrator. Specifically, petitioner alleges that since the Administrator’s approval of the plan, low-sulfur coal has become too scarce and expensive to obtain; rehable and satisfactory sulfur dioxide removal equipment that would enable it to comply with the plan’s requirements simply has not been devised; the installation of the unsatisfactory equipment that is available would cost over $500 million, a sum impossible to obtain by bonds that are contingent on approval by regulatory bodies and public acceptance; and, even if the financing could be obtained, the carrying, operating, and maintenance costs of over $120 million a year would be prohibitive.2 Petitioner further alleges that recent evidence has disclosed that sulfur dioxide in the ambient air is not the hazard to public health that it was once thought to be, and that compliance with the sulfur regulation in the Missouri plan is not necessary to the attainment of national primary and secondary ambient air standards in the St. Louis area.

At the risk of civil and criminal penalties enforceable by both the State and Federal Governments, as well as possible citizens’ suits, 42 U. S. C. §§ 1857c-8, 1857h-2, petitioner is being required either to embark upon the task of installing allegedly unreliable and prohibitively expensive equipment or to shut down. Yet the present Act permits neither the Administrator, in approving the state plan, nor the courts, in reviewing that approval under § 307 of the Act, 42 U. S. C. § 1857h-5, even to consider petitioner’s allegations of infeasibility.

Environmental concerns, long neglected, merit high priority, and Congress properly has made protection of *272the public health its paramount consideration. See S. Rep. No. 91-1196, pp. 2-3 (1970). But the shutdown of an urban area’s electrical service could have an even more serious impact on the health of the public than that created by a decline in ambient air quality. The result apparently required by this legislation in its present form could sacrifice the well-being of a large metropolitan area through the imposition of inflexible demands that may be technologically impossible to meet and indeed may no longer even be necessary to the attainment of the goal of clean air.

I believe that Congress, if fully aware of this Draconian possibility, would strike a different balance.

Opinion for the Court filed by Circuit Judge RANDOLPH.

RANDOLPH, Circuit Judge:

These are consolidated petitions for review of a final rule issued by the Environmental Protection Agency under the Clean Air Act (“CAA”). The rule is aimed at reducing ozone pollution in the northeastern United States. It requires the twelve states in the region and the District of Columbia to adopt what is essentially California’s vehicle emission program. Opposing the rule, and appearing here as petitioners, are the Commonwealth of Virginia, and three associations representing automobile manufacturers and dealers. Intervening to defend the rule are the Commonwealth of Massachusetts, the States of New York, Connecticut, Rhode Island and Vermont, the City of New York, and two associations.

Petitioners believe EPA’s rule is unsupported by the record, contrary to the statute, and constitutionally defective. The evidence regarding ozone pollution, they maintain, does not support EPA’s demand for region-wide emission reductions and, in any event, the means EPA has chosen for realizing those reductions — mandating that these states prohibit the sale of new cars that do not satisfy California’s standards — is something Congress has barred the Agency from imposing. A majority of the governors of the twelve northeastern states recommended, over the objection of Virginia and other states, that EPA impose this solution. The group operated as a regional commission under a new section of the Clean Air Act, a section petitioners say is unconstitutional because it gave the commission significant legislative and executive powers without regard to the Appointments Clause of the Constitution, the non-delegation doctrine, the Joinder and Compact Clauses and other constitutional limitations. Furthermore, in petitioners’ view EPA’s interpretation of the Act renders it unconstitutional under the Tenth Amendment because it compels the states to enact and administer a federal regulatory program.

I

A

EPA believed that if the states in the Northeast passed legislation adopting the California ear program this would help alleviate the ozone hazard. Ozone (03) in the upper atmosphere was not the problem. There it absorbs harmful ultraviolet rays. Ozone at ground level is another matter entirely. It is one of the primary constituents of smog. Ozone’s three-atom arrangement is highly unstable: the third oxygen atom, in a process called oxidation, has an aggressive tendency to react with whatever substance is available. Ozone’s high reactivity, evident in the stratosphere where ozone reacts with chlorofluorocarbons, has harmful effects at ground level. Much of the ozone inhaled *1400reacts with sensitive lung tissues, irritating and inflaming the lungs, and causing a host of short-term adverse health consequences including chest pains, shortness of breath, coughing, nausea, throat irritation, and increased susceptibility to respiratory infections. Final Rule, 60 Fed.Reg. 4712, 4712-13 (1995). “The mechanisms of ozone-induced impairment of lung function are only partly understood,” but “the effects of a single exposure to” ozone “are reversible and last up to 48” hours. Editorial, Ozone: Too Much in the Wrong Place, The Lancet, July 27, 1991, at 221. On the other hand, “[c]hronic effects that have resulted from recurrent seasonal exposure to ozone have been studied only to a limited degree. Most of the current evidence is derived from animal responses to chrome ozone exposure.” Committee ON TROPOSPHERIC OZONE FORMATION and Measurement, National Research Council, Rethinking the Ozone PROBLEM in Urban and Regional Air Pollution 33 (1991) [hereinafter National Research Council]. In addition to its direct effect on humans, excessive ozone can also damage forests and food crops.

Exactly how ozone is created and transported in the lower atmosphere, and how it decays, is a matter of extreme complexity. Ozone is not a direct pollutant. Vehicles do not emit it, and it does not billow out of smokestacks. Instead, it is formed mostly from the mixture of two chemical precursors emitted by automobiles and industry: nitrogen oxides (NOx) and a large group of hydrocarbon pollutants called volatile organic compounds (VOCs).1 National Research Council, supra, at 153. These precursors cook in the sun — they cook best in a “sea” of warm stagnant air — and produce ozone through a complex chain of reactions. The creation of ozone can thus be seen as a seasonal phenomenon, with concentrations peaking in the summer, and as a diurnal occurrence, with concentrations peaking during the afternoon and falling during the night. The precursor- and ozone-laden air slowly moves downwind, and as the air mass moves, ozone levels often continue to increase, in part because the ozone has more time to develop, in part because the air mass picks up more precursors along the way. Ultimately, this process can bring high ozone levels to areas hundreds of miles downwind of the pollution sources.- See id. at 100. Therein lies the source of much of the difficulty in controlling it.

At the moment, dozens of areas throughout the United States have not attained the national ambient air quality standard for ozone. The standard permits a maximum one-hour average ozone concentration of 0.12 parts per million. 40 C.F.R. § 50.9. If the ambient air exceeds this standard more than once per year averaged over three years, the area is considered in “nonattainment.” Id. The ozone season of 1994 was a good one; EPA estimated that there was twelve percent less ozone than in 1985. EPA Air Quality Trends Brochure 1994 (EPA-454/F-95-003) (1995). But the warmer 1995 season was worse, with national ozone concentrations increasing four percent from 1994. EPA Air Quality Trends Brochure 1995 (EPA-454/F-96-008) (1996). According to press reports, New Jersey and Maryland each exceeded the ambient air standard fourteen times. See Northeast Ozone Problems Continue, Show Long-Term Gain, Octane WK., Oct. 9,1995, at 2.

Because of the ozone transport phenomenon, an area in nonattainment may be unable to do much about it. Even if the state implements the strictest practicable control measures, it might still fail to satisfy the ambient air standard. Across the country there are what one might call interstate ozone transport regions: around Chicago; in the region from Cleveland to Erie, Pennsylvania; in the Texas-Louisiana area; and in the northeast from Virginia to Maine, an area that became subject to special legislation in 1990.

*1401B

The “Northeast Ozone Transport Region,” defined in the 1990 amendments to the Clean Air Act, consists of Connecticut, Delaware, Maine, Maryland, Massachusetts, New Hampshire, New Jersey, New York, Pennsylvania, Rhode Island, Virginia, Vermont, and the District of Columbia.2 To decrease ozone pollution in this “Region,” EPA promulgated the rule we have before us. Each of these states had an implementation plan, adopted after public hearings, containing measures to control air pollution, and approved by EPA under section 110. CAA § 110(a)(1), 42 U.S.C. § 7410(a)(1). EPA’s rule declared all of these implementation plans “substantially inadequate” and therefore in need of revision. Final Rule, 60 Fed.Reg. 4712, 4736 col. 2 (1995). The plans were inadequate, according to EPA, because nitrogen oxide and volatile organic compound emissions had to be reduced by 50% to 75% from a 1990 baseline “throughout” this densely populated Region to attain the national ambient air quality standard for ozone in “serious and severe areas.” Id. at 4720 col. 1. EPA acknowledged that it was “enormously complicated to determine which reductions” were needed where. Id. col. 2. But, it said, “wind trajectory data” supported the following conclusions. Id. col. 3. During the summer, ozone and emissions of precursors in the Washington, D.C. area contribute to the ozone problem in Baltimore, which lies to the northeast. Baltimore and the rest of Maryland contribute to ozone pollution in the Philadelphia area, which contributes to the problem in New York, which contributes to the problem in Boston, and so on up the Eastern seaboard from one metropolitan area to another. Id. at 4720-21. (Of course the northeastern section of the country is not some isolated island; and there are indications that western Pennsylvania receives ozone from Ohio and West Virginia.)

Congress, too, had found the entire subject very complex. In the 1990 amendments it ordered EPA, “in conjunction with the National Academy of Sciences, [to] conduct a study on the role of ozone precursors in tropospheric ozone formation and control. The study shall examine the roles of NOx and VOC emission reductions, the extent to which NOx reductions may contribute (or be counterproductive) to achievement of attainment in different nonattainment areas....” CAA § 185B, 42 U.S.C. § 7511f. One part of the study, completed in 1991, found that reducing NOx emissions can sometimes increase, rather than decrease, ambient ozone levels. See National RESEARCH Council, supra, at 11. “This seemingly contradictory prediction, that lowering NOx can, under certain conditions, lead to increased ozone, results from the complex chemistry involved in ozone formation in VOC-NOx mixtures_” Id. at 167.

Nonetheless, having declared the state plans inadequate, EPA required the states to revise their plans and enact a Low Emission Vehicle program, or LEV as the parties call it. In EPA’s words, the final rule orders “all of the northeastern states to adopt the. California car program to reduce significantly the pollution emitted by new cars and' light-duty trucks.” Final Rule, 60 Fed.Reg. at 4713 col. 1. To appreciate the rationale of this order one must understand that there are two, and only two, permissible sets of regulations limiting emissions from new cars sold in the United States. There are the California regulations; and there are the federal regulations, which preempt the laws of the other 49 states. CAA §§ 202, 209(a), 42 U.S.C. §§ 7521, 7543(a). The California standards — the “Low Emission Vehicle” standards — are considerably more restrictive than the federal standards and include ceilings on motor vehicle emission of nitrogen *1402oxides, see 13 Cal. Admin.Code § 1960.1(f)(2), (g)(1), and volatile organic compounds, see 13 Cal. Admin.Code § 1960.1(f)(2), (g)(1), (g)(2), the two chief ozone precursors.

C

In finding the twelve state plans inadequate, and in requiring the states to adopt the California solution, EPA acted on the recommendation of the “Northeast Ozone Transport Commission,” a body established by section 184 of the Act. CAA § 184, 42 U.S.C. § 7511c. As part of the 1990 amendments, section 184 set up a panel consisting of the governors of the twelve states (and the Mayor of the District of Columbia) or their delegates. This panel, or Ozone Commission as we shall call it, was to develop, through majority vote, proposals for additional control measures for ozone pollution in the Region “necessary to bring any area in such region into attainment.” Id. § 7511c(e)(l).3

After EPA receives a Commission proposal, EPA must “publish in the Federal Register a notice stating that the recommendations are available and provide an opportunity for public hearing within 90 days beginning on the receipt date”; “commence a review of the recommendations to determine whether the control measures in the recommendations are necessary to bring any area into attainment” with national standards; and “consult with members of the commission” and take into account data and comments received during the notice and comment process. Id. § 7511c(c)(2), (c)(3).

From the time of receiving the Commission’s recommendations, EPA has nine months: to “determine whether to approve, disapprove,” or approve in part and disapprove in part, the recommendation; to “notify the commission in writing” of the Administrator’s determination; and to “publish such determination in the Federal Register.” Id. § 7511c(c)(4). Absent complete approval, EPA must explain “why any disapproved additional control measures are not necessary to bring any area in such region into attainment ... or are otherwise not consistent with” the remainder of the Act, and must recommend “actions that could be taken by the commission to conform the disapproved portion of the recommendations to the requirements” of section 184. Id. § 7511c(c)(4)(i) & (c)(4)(H).

If EPA approves the recommendation, it must declare each state’s implementation plan inadequate and it must order the states to include the approved control measures in their revised plans pursuant to section 11000(5). CAA § 184(c)(5), 42 U.S.C. § 7511c(c)(5). Failure to heed EPA’s order unleashes a panoply of sanctions upon the noncomplying state. We will have more to say on the subject of sanctions later in the opinion.

EPA’s final rule in this case resulted from the section 184 process just described. In August 1993, Maine, Maryland, and Massachusetts petitioned the Ozone Commission to adopt a recommendation calling for the application of the California Low Emission Vehicle program throughout the Region. See Notice of Availability, 59 Fed.Reg. 12,914, 12,915 col. 1 (1994). After holding several “public forums,” and a hearing, the Commission voted on February 1, 1994, “to recommend that EPA mandate the California” vehicle program “throughout the” Region. Id. A minority of Ozone Commission members— those from Virginia, Delaware, New Jersey, and New Hampshire — voted against the recommendation. See id. On February 10, *14031994, the Commission submitted the recommendation to EPA Id. at 12,914 col. 1.

In its final rule, EPA approved the Commission’s recommendation. 60 Fed.Reg. at 4713 col. 1. Therefore, as section 184(c)(5) required, EPA found “that the State Implementation Plans” for the twelve states in the Region “are substantially inadequate to comply with the requirements of section 110(a)(2)(D) of the Clean Air Act, 42 U.S.C. § 7410(a)(2)(D), and to mitigate adequately the interstate pollutant transport described in section 184 of the Clean Air Act, 42 U.S.C. § 7511c, to the extent that they do not provide for emission reductions from new motor vehicles in the amount that would be achieved by the” Low Emission Vehicle program. Final Rule, 60 Fed.Reg. at 4736 col. 2 (to be codified at 40 C.F.R. § 51.120(a)).

D

EPA purported to offer the states an alternative to enacting the California program. Instead of restricting the type of new cars sold within its jurisdiction, a state may choose its own mix of programs. To be an acceptable “Substitute Program,” however, it must reduce ozone precursors anywhere from 3.5 to 6.5 times more than would the California program. See infra part II.A4

II

Of petitioners’ many challenges , to EPA’s rule, one is aimed at something known in Clean Air Act parlance as a “SIP call”— EPA’s declaration that a state’s implementation plan is substantially inadequate and must be revised. Virginia maintains that the record does not support the SIP call to it. If Virginia is correct, EPA’s regulation requiring Virginia to adopt the California car program would fall. However, the issues raised by the other petitioners — the automobile manufacturers and dealers — reach beyond Virginia. If the states implement the California program in response to EPA’s rule, these petitioners claim their members will be adversely affected throughout the Northeast Region as new car price increases take effect.5 Furthermore, it is not clear whether EPA would have issued a SIP call to any state if EPA knew it lacked a legal basis for requiring the California program. We therefore think it best to start our analysis with questions other than the SIP call, questions relating mainly to EPA’s statutory authority.

We will discuss those questions in this order and explain why we have reached the following conclusions.. First, did EPA condition its approval of a state’s revision of its *1404implementation plan on the state’s adoption of the California program? We think the answer is yes. Second, does section 110 give EPA the authority to condition approval of a state’s plan on the state’s adoption of control measures EPA has chosen? We conclude that section 110 does not give EPA this authority. Third, does section 184 authorize EPA to so condition its approval of plan revisions? We believe the answer is yes. Fourth, did other provisions of the Clean Air Act nevertheless bar EPA from ordering states to enact the California program? We hold that EPA was so barred and that this portion of its final rule is therefore invalid.

A

Did EPA condition approval of a state’s revised implementation plan on the state’s adoption of the California ear program? EPA tells us it did not because it offered the states a choice — they could either enact the California program or submit a “Substitute Program.” It turns out, however, that EPA’s alternative is no alternative at all.

Having determined that, at a minimum, 50 percent reductions of each precursor are necessary to bring each northeastern state into ozone attainment, 60 Fed.Reg. at 4718 col. 3, EPA required any Substitute Program to reduce precursor emissions by the difference between that 50 percent and the percentage reduction “achievable through implementation of all of the Clean Air Act-mandated and potentially broadly practicable control measures,” id. at 4737 col. 2 (to be codified at 40 C.F.R. § 51.120(d)(1) & (d)(2)). The phrase “potentially broadly practicable control measures” refers to other options — “those control measures that could potentially render the” California program “unnecessary in the” Northeast. Supplemental Notice, 59 Fed. Reg. at 48,678 col. 2. EPA has calculated that “potentially broadly practicable control measures” yield a 35.9 percent reduction in nitrogen oxides6 and a 37 percent reduction in volatile organic compounds. Final Rule, 60 Fed.Reg. at 4722. Since none of those options (taken individually or cumulatively) could produce the necessary 50 percent reduction for precursors, EPA takes the position that all the measures are necessary, and presupposes that those measures will be included in state implementation plans submitted sometime in the future. Id. at 4723 col. 3.

Thus, a state that chooses not to adopt the California program must legislate a state implementation plan containing measures to make up this “shortfall” — the difference between the 50 percent reductions EPA says is necessary and the 35.9 percent and 37 percent that it has already targeted for nitrogen oxides and volatile organic compounds, respectively. Virginia or any other state seeking to implement a program not involving California vehicles, therefore, must pass legislation or issue regulations or orders reducing nitrogen oxide production in the year 2005 by 14.1 percent (50 percent minus 35.9 percent), and volatile organic compound production by 13 percent (50 percent minus 37 percent).

In contrast, the California program, by 2005, reduces' nitrogen oxides by 4 percent, and volatile organic compounds by 2 percent. Supplemental Notice, 59 Fed.Reg. at 48,683 col. 1. States adopting the California program need legislate nothing more.7 Virginia, *1405therefore, must reduce nitrogen oxides 3.5 times (14.1%/4%) and volatile organic compounds 6.5 times (13%/2%) more than do states that pass legislation incorporating the California program, and Virginia must do so by means EPA has already concluded are not practicable.8 In sum, the Substitute Program is expected to solve a shortfall EPA itself has no idea how to remedy, a shortfall states enacting the California program are not expected to make up.

Of course, only a very foolish state would see EPA’s offer to accept this Substitute Program as a real alternative. EPA conceded as much. No state, EPA recognized, “would seriously entertain unreasonable or impracticable measures to adopt in place of the” California program “in order to achieve the necessary emissions reductions.” Supplemental Notice, 59 Fed.Reg. at 48,672 col. 2. As a practical matter, then, EPA “require[d] all the northeastern states to adopt the California car program.” Final Rule, 60 Fed.Reg. at 4713 col. 1. And that is how we shall treat the rule.

B

Does section 110 give EPA the authority to condition approval of a state’s plan on the state’s adoption of control measures EPA has chosen? EPA’s discovery that section 110 gave it this authority came late in the rulemaking proceedings. The Ozone Commission, acting pursuant to section 184, started the ball rolling. It forced ozone pollution in the Northeast to the forefront of EPA’s agenda. In EPA’s Notice of Proposed Rulemaking, the agency therefore naturally treated section 184 as the sole source of its authority to approve or disapprove the Commission’s recommendation. See 59 Fed. Reg. 21,720, 21,720-21 (1994).

The agency shifted its position after automobile manufacturers and dealers began questioning the constitutionality of section 184. See Supplemental Notice, 59 Fed.Reg. at 48,670 col. 1 & n.ll. In response to the proposed rule, some argued that EPA deference to the Ozone Commission’s technical and policy judgments would violate the Appointments Clause of the Constitution9 and the nondelegation doctrine. Commenters also complained that the Joinder and Compact Clauses10 prevented a multi-state alliance such as the Ozone Commission from exercising significant power. And they maintained that the proposed rule forcing the California program on the states would violate the Tenth Amendment because it would “‘commandeer’ state governments into the service of federal regulatory purposes, and would for this reason be inconsistent with the Constitution’s division of authority between federal and state governments.” New York v. United States, 505 U.S. 144, 175, 112 S.Ct. 2408, 2428, 120 L.Ed.2d 120 (1992).

In response to these comments, EPA issued a süpplemental notice stating that the arguments against section 184’s constitutionality lacked “merit,” and that “in any case, given EPA’s independent authority under section 110(k)(5), any constitutional question *1406regarding the validity of section 184 would not affect the validity of the” proposed rule. Supplemental Notice, 59 Fed.Reg. at 48,670 col. 1. And so, not only section 184, but also section 110 supposedly provided the legal basis for EPA’s action.

C

Enacted more than a quarter of a century ago, section 110 has gone through many changes, but its basic structure has survived. The provision has been the subject of dozens of appellate court decisions. Yet we are aware of no ease (EPA has cited none) supporting the proposition EPA now urges upon us, namely, that under section 110 EPA may condition approval of a state’s implementation plan on the state’s adopting a particular control measure, here the California Low Emission Vehicle program. Despite the volumes of legislative history on the Clean Air Act and its amendments, EPA has offered no committee reports, no statements on the floor of either house, suggesting that anyone in Congress ever shared its view of section 110. Nor has EPA given us reason to believe that it has followed any consistent, longstanding practice of using section 110 to force states to adopt control measures of EPA’s choosing. The final rule provides no explanation of why EPA read section 110 as it did. The rule contains only this bare conclusion: “EPA disagrees with comments claiming that EPA lacks” authority “because section 110 does not authorize EPA to require states to adopt specific measures.... ” Final Rule, 60 Fed.Reg. at 4717 col. 1. EPA’s brief offers little more. We are told the “plain language” of the statute gives EPA this authority, and even if the language is not so plain, EPA’s reading is reasonable and deserves our deference. Final Brief for Respondent at 34. We are not persuaded, and therefore do not reach Virginia’s contention that EPA’s interpretation would render section 110 unconstitutional, in violation of the Tenth Amendment.

Congress added section 110 to the Clean Air Act in 1970. See Clean Air Amendments of 1970, Pub.L. No. 91-604, § 4, 84 Stat. 1676,1680-81 (1970). It has remained one of the key provisions of the Act. The Act authorizes EPA to promulgate national ambient air quality standards for ozone and five other pollutants. CAA §§ 108 & 109, 42 U.S.C. §§ 7408 & 7409. Areas that do not meet the minimum level of air quality mandated by these national standards are considered to be “nonattainment areas.” CAA §§ 107(d) & 171(2), 42 U.S.C. §§ 7407(d) & 7501(2). The degree of nonattainment is classified as marginal, moderate, serious, severe, or extreme. CAA § 181(a), 42 U.S.C. § 7511(a).

If a state has an area within it that EPA has classified as being in nonattainment with respect to ozone (or one of the five other regulated pollutants), the state must devise and implement a “state implementation plan” (“State Plan”). Section 110 governs the interplay between the states and EPA with respect to the formulation and approval of such State Plans. 42 U.S.C. § 7410. The basic procedure is that “each state determines an emission reduction program for its nonattainment areas, subject to EPA approval, within deadlines imposed by Congress.” Natural Resources Defense Council, Inc. v. Browner, 57 F.3d 1122, 1123 (D.C.Cir.1995).

Should a state fail to submit an implementation plan, or should its plan fail to provide the required reductions in air pollution, certain penalties — some mandatory, others at EPA’s discretion — may follow. See generally Commonwealth of Virginia v. United States, 74 F.3d 517, 520 (4th Cir.1996); Natural Resources Defense Council, 57 F.3d at 1123-25 & n. 5. The noncomplying state may, for instance, be prevented fipm spending federal highway money in nonattainment areas. See CAA §§ 110(m), 176(c) & 179(b)(1); 42 U.S.C. §§ 7410(m), 7506(c) & 7509(b)(1). This sanction becomes mandatory if the state fails to implement an adequate State Plan within 24 months of EPA’s finding that the state’s proposed plan is deficient. CAA § 179(b)(1), 42 U.S.C. § 7509(b)(1). At that same point, EPA must impose a “federal implementation plan” (“Federal Plan”) on those areas of the state in nonattainment. CAA § 110(c), 42 U.S.C. § 7410(c). The Federal Plan “provides an additional incentive for state compliance because it rescinds state authority to make the many sensitive and policy choices that a pollution control' *1407regime demands.” Natural Resources Defense Council, 57 F.3d at 1124.

Section 110(a)(2), the' predecessor to the current section 110(k)(5), spelled out the extent of EPA’s authority over state implementation plans within this structure.11 When enacted in 1970, the relevant portion of section 110(a)(2) read:

The Administrator shall, within four months after the date required for submission of a plan under paragraph (1), approve or disapprove such plan, or each portion thereof. The Administrator shall approve such plan, or any portion thereof, if he determines that it was adopted after reasonable notice and hearing and that—

* * *

(H) it provides for revision, after public hearings, of such plan (i) from time to time as may be necessary to take account of revisions of such national primary or secondary ambient air quality standard or the availability of improved or more expeditious methods of achieving such primary or secondary standard; or(ii) whenever the Administrator finds on the basis of information available to him that the plan is substantially inadequate to achieve the national ambient air quality primary or secondary standard which it implements.

42 U.S.C. § 1857e-5(a)(2)(H) (1970) (emphasis added). At the time, section 110(a)(3) required EPA to review a state’s revisions to its plan — which is what EPA is calling for here — by the same criteria EPA used to judge the original plan. 42 U.S.C. § 1857c-5(a)(3) (1970). In 1970 as now, if a state failed to submit a plan, or if the plan did not satisfy section 110(a)(2), EPA had a non-discretionaiy duty to implement the Act through a federal implementation plan. CAA § 110(c), 42 U.S.C. § 1857e-5(c) (1970).

In 1975, the Supreme Court analyzed section 110’s “division of responsibilities” between the states and the federal government. Train v. Natural Resources Defense Council, 421 U.S. 60, 79, 95 S.Ct. 1470, 1482, 43 L.Ed.2d 731 (1975). The narrow issue in Train was whether states could use section 110(a)(3) to seek “revisions” of State Plans in order to ease limits on a single pollution source, in essence circumventing section 110(f), which restricted the circumstances under which such a “variance” could be granted. A state could, the Court held, use section 110(a)(3) for such a purpose so long as the revised plan would still enable the state to attain and maintain national ambient standards. Train, 421 U.S. at 72, 95 S.Ct. at 1478.

A “broader issue,” the Court thought, was at stake in the case — namely, “whether Congress intended the States to retain any significant degree of control over the manner in which they attain and maintain national standards.” Id. at 78, 95 S.Ct. at 1481. The Act expressly gave the states initial responsibility for determining the manner in which air quality standards, were to be achieved. See id. at 64, 95 S.Ct. at 1474-75. Section 107(a) of the Act read then, as it does now: “Each State shall have the primary responsibility for assuring air quality within the entire geographic area comprising such State by submitting an implementation plan which will specify the manner in which national primary and secondary ambient air quality standards will be achieved and maintained within each air quality control region in such State.” CAA § 107(a), 42 U.S.C. § 7407(a) (emphasis added). In light of section 107(a), the Court construed section 110:

The Act gives the Agency no authority to question the wisdom of a State’s choices of emission limitations if 'they are part of a plan which satisfies the standards of § 110(a)(2), and the Agency may devise and promulgate a specific plan of its own *1408only if a State fails to submit an implementation plan which satisfies those standards. § 110(e). Thus, so long as the ultimate effect of a State’s choice of emission limitations is compliance with the national standards for ambient air, the State is at liberty to adopt whatever mix of emission limitations it deems best suited to its particular situation.

Train, 421 U.S. at 79, 95 S.Ct. at 1482. The Supreme Court repeated this interpretation in Union Electric Co. v. EPA, 427 U.S. 246, 96 S.Ct. 2518, 49 L.Ed.2d 474 (1976): section 110 left to the states “the power to determine which sources would be burdened by regulations and to what extent.” Id. at 269, 96 S.Ct. at 2531. To be sure, if EPA rejected a State Plan because it would not achieve or maintain ambient air quality standards, EPA could promulgate a federal implementation plan. But even then, as EPA admitted in confessing error in the Supreme Court, section 110 did not permit the agency to require the state to pass legislation or issue regulations containing control measures of EPA’s choosing. EPA v. Brown, 431 U.S. 99, 103, 97 S.Ct. 1635, 1636-37, 52 L.Ed.2d 166 (1977) (per curiam).

In 1977 Congress again amended the Act, changing some parts of section 110.12 But these changes did not modify the “division of responsibilities” Train had discerned in the Act. EPA’s later administrative efforts to alter the balance were therefore firmly rebuffed. Bethlehem Steel Corp. v. Gorsuch, 742 F.2d 1028 (7th Cir.1984), held that EPA could not, under the guise of partially approving a state implementation plan, render the plan more stringent than the state intended. If EPA wanted to impose stricter regulations, the Clean Air Act gave it the option of implementing a federal plan. The court explained:

[T]he Clean Air Act creates a partnership between the states and the federal government. The state proposes, the EPA disposes. The federal government through the EPA determines the ends — the standards of air quality — but Congress has given the states the initiative and a broad responsibility regarding the means to achieve those ends through state implementation plans and timetables of compli-ance_ The Clean Air Act is an experiment in federalism, and the EPA may not run roughshod over the procedural prerogatives that the Act has reserved to the states, ... especially when, as in this case, the agency is overriding state policy.

Id. at 1036-37. Florida Power & Light Co. v. Costle, 650 F.2d 579 (5th Cir.1981), is to the same effect. EPA’s “attempting to require Florida to include” a particular provision in its State Plan was “clearly an abuse of discretion; it is agency action beyond the Congressional mandate,” action that would “usurp state initiative in the environmental realm,” and “disrupt the balance of state and federal responsibilities that undergird the efficacy of the Clean Air Act.” Id. at 587, 589.

Thus, as section 110 stood in 1975 when the Supreme Court decided Train and as it stood after the 1977 amendments, the provision did not confer upon EPA the authority to condition approval of Virginia’s implementation plan, or the plan of any other state, on the state’s adoption of a specific control measure. EPA “identifies the end to be achieved, while the states choose the particular means for realizing that end.” Air Pollution Control Dist. v. USEPA 739 F.2d 1071, 1075 (6th Cir.1984). The validity of EPA’s argument that it now has broader authority depends, therefore, on whether the 1990 Clean Air Act Amendments eliminated the “liberty” every state had under the earlier *1409legislation “to adopt whatever mix of emission limitations it deems best suited to its particular situation.” Train, 421 U.S. at 79, 95 S.Ct. at 1482.

The 1990 amendments were the most comprehensive since 1970, but the changes to section 110, at least as they concern EPA’s approval of State Plans, were predominantly of syntax, not substance. Section 110(a)(2)(H)(ii), as amended, requires each State Plan to “provide for revision” not only when EPA finds the plan “substantially inadequate to attain” national ambient standards, but also when EPA finds that the plan does not “otherwise comply with any additional requirements established under this chapter.” CAA § 110(a)(2)(H)(ii), 42 U.S.C. § 7410(a)(2)(H)(ii).

Also in the 1990 law, Congress repealed section 110(a)(3)(A), a provision requiring the Administrator to approve a State Plan revision meeting the (a)(2) criteria. The language of the repealed subsection was moved to a new subsection (k), so that the new section 110(k)(3) imposed the same obligation upon the Administrator to approve plans satisfying the (a)(2) requirements: “In the case of any” State Plan submission or revision, “the Administrator shall approve such sub-mittal as a whole if it meets all of the requirements of this chapter.” Id. § 7410(k)(3).

The 1990 amendments added subsection (k)(5), on which EPA relies in this ease. Like the original section 110(a)(2)(H)(ii), this new section 110(k)(5) gave EPA authority to require states to revise their plans if they have failed to attain or maintain national ambient air quality standards. 42 U.S.C. § 1857c-5(a)(2)(H)(2) (1970); 42 U.S.C. § 7410(k)(5) (1994). So far as EPA’s authority is concerned, the new subsection merely duplicated the language in the 1970 version of section 110(a)(2)(H)(ii). The first two sentences of section 110(k)(5) now read:

Whenever the Administrator finds that the applicable implementation plan for any area is substantially inadequate to attain or maintain the relevant national air quality standard, to mitigate adequately the interstate pollution transport described in [CAA sections 176A or 184] or to otherwise

comply with any requirement of this Act, the Administrator shall require the State to revise the plan as necessary to correct such inadequacies. The Administrator shall notify the State of the inadequacies, and may establish reasonable deadlines (not to exceed' 18 months after the date of such notice) for the submission of such plan revisions.

42 U.S.C. § 7410(k)(5).

The main differences between this portion of subsection (k)(5) and the original section 110(a)(2)(H)(ii) are that EPA may set the timetable for a state to revise its plan, and that a plan’s failure “to mitigate adequately the interstate pollution transport described in” sections 176A or 184 may trigger the need for revision. Neither of these differences, however, amounts to a new grant of authority to EPA to require states to insert in their plans control measures EPA has selected. The original act, interpreted in Train v. Natural Resources Defense Council, 421 U.S. 60, 95 S.Ct. 1470, 43 L.Ed.2d 731 (1975), permitted EPA to call upon a state to revise its plan whenever the Administrator found the plan substantially inadequate. Section 110(k)(5) does the same.

EPA apparently thinks the “as necessary” language in section 110(k)(5) altered the division of responsibilities between the states and the agency. We suppose the idea is that because section 110(k)(5) empowers EPA to “require the State to revise the plan as necessary to correct” inadequacies, it empowers EPA to require the state to include particular control measures in the revised plan. There is nothing to this.

The Supreme Court’s decision in Train provides the starting point, and is itself the endpoint for interpreting section 110, unless an “intervening development of the law has ‘removed or weakened the conceptual underpinnings from the prior decision, or ... the later law has rendered the decision irreconcilable with competing legal doctrines or policies.’ ” Neal v. United States, - U.S. -, -, 116 S.Ct. 763, 768-69, 133 L.Ed.2d 709 (1996) (quoting Patterson v. McLean Credit Union, 491 U.S. 164, 173, 109 S.Ct. 2363, 2370-71, 105 L.Ed.2d 132 (1989)). *1410Congress’ insertion of “as necessary” in section 110(k)(5) does not affect the underpinnings of Train, nor is the Train interpretation irreconcilable with those two words. All this additional language does is keep EPA within bounds. If a state plan is inadequate — that is, if the state is not achieving an ambient air quality standard — EPA can call only for revisions “as necessary” to correct that problem. EPA cannot, in other words, require the state to go through a wholesale revision of its entire plan. In that respect, section 110(k)(5) complements section 110(a)(2)(H)(ii), the section requiring State Plans to contain procedures for revisions. We know, for the reasons spelled out above, that Congress did not give EPA authority to choose the control measures or mix of measures states would put in their implementation plans. And we can think of no reason why Congress, merely by inserting the words “as necessary,” would have meant to hand over that authority to EPA when it calls upon states to revise their implementation plans.13 We would have to see much clearer language to believe a statute allowed a federal agency to intrude so deeply into state political processes. See Gregory v. Ashcroft, 501 U.S. 452, 463, 111 S.Ct. 2395, 2402, 115 L.Ed.2d 410 (1991).

We therefore hold that the 1990 amendments did not alter the division of responsibilities between EPA and the states in the section 110 process. It was with this understanding that we recently summarized the statutory system: “The states are responsible in the first instance for meeting the” national ambient standards “through state-designed plans that provide for attainment, maintenance, and enforcement of the” national standards “in each air quality control region. Thus, each state determines an emission reduction program for its nonattainment areas, subject to EPA approval, within deadlines imposed by Congress.” Natural Resources Defense Council, Inc. v. Browner, 57 F.3d 1122, 1123 (D.C.Cir.1995).

Because section 110 does not enable EPA to force particular control measures on the states, EPA’s authority to promulgate the rule under review must be derived from section 184 alone.

D

Does section 184 give EPA the authority to condition approval of a state’s plan on the state’s adoption of particular control measures? The language of this statute, unlike section 110, answers with an emphatic yes. By a majority vote, the Ozone Commission recommends “additional control measures” for all or part of the Region. CAA § 184(c)(1), 42 U.S.C. § 7511c(e)(l). If EPA approves the recommendation, section 184(c)(5) requires it to issue a “finding” that the implementation plans of the affected states are inadequate. “Such finding shall require each such State to revise its implementation plan to include the approved additional control measures within one year after the finding is issued.” Id. We think the only possible reading of section 184(c)(5) is that a state must revise its plan to include whatever “additional control measures” EPA has approved, or it must suffer the consequences of losing its federal highway funds, of having new factories and other “major stationary sources” of air pollution restricted, and of having the federal government impose an implementation plan in the state. See Commonwealth of Va. v. Browner, 80 F.3d 869, 881-83 (4th Cir.1996).

Virginia and the other petitioners do not quarrel with this reading of section 184. But, they say, section 184 is unconstitutional in light of the role of the Ozone Commission. We would not have to reach the constitutional questions if, as Virginia also argues, sections 202 and 177 of the Clean Air Act barred EPA from ordering states to implement the California car program. It is an old principle that if a case may be decided on statutory grounds “it will be unnecessary, and consequently improper, to pursue any inquiries, which would then be merely speculative, respecting the power of Congress in the case.” Cohens v. Virginia, 19 U.S. (6 Wheat.) 264, 441, 5 L.Ed. 257 (1821) (Marshall, C.J.); Ash-*1411wander v. Tennessee Valley Auth., 297 U.S. 288, 347, 56 S.Ct. 466, 483, 80 L.Ed. 688 (1936) (Brandéis, J., concurring). And so we turn to those two sections.

E

Does section 177, read together with section 202, forbid EPA from conditioning its approval of a state’s implementation plan on the state’s adoption of the California program to limit motor vehicle emissions? We hold that this provision did so bar EPA and that, in this respect, the final rule is invalid.

Section 202 of the Clean Air Act, codified at 42 U.S.C. § 7521, regulates emission standards for new motor vehicles or new motor vehicle engines. The standards, covering all air pollutants that may endanger public health or welfare, preempt the laws of every state except California. CAA § 209(a) & (b), 42 U.S.C. § 7543(a) & (b). Why Congress exempted California, and how, is explained in Motor Vehicle Manufacturers Ass’n v. New York State Department of Environmental Conservation, 17 F.3d 521, 524-28 (2d Cir.1994). We will not go into the subject here.

With respect to the federal “Phase I” standards applicable to all the states except California, subparagraph (b)(3)(C) of section 202 provides: “It is the intent of Congress that the numerical emission standards specified in subsections (a)(3)(B)(ii), (g), (h), and (i) of this section shall not be modified by the Administrator after November 15, 1990, for any model year before the model year 2004.”14 CAA § 202(b)(3)(C), 42 U.S.C. § 7521(b)(3)(C). Section 202(b)(3)(C) thus prevents EPA from prescribing, before the model year 2004, emissions standards for motor vehicles more restrictive than those already written into the statute. Accord id. § 202(f)(1), (i)(3)(A)(ii), (i)(3)(B)(ii), (i)(3)(C), (i)(3)(E). The proviso just quoted resolved a debate in the Senate about whether there should be any future rounds of heightened emission standards. See, e.g., 136 Cong. Reg S250 (daily ed. Jan. 24, 1990) (statement of Sen. Baucus), reprinted in Legislative His-TOKY OF THE CLEAN AlR ÁCT AMENDMENTS OF 1990, at 4941 (1993) [hereinafter 1990 Legislative History]. Senator Wirth, one of the sponsors of the bill, explained: ‘We do not know if the second round is going to be needed. This compromise was reached. Let us have a jump ball on the situation. Defer the question until the late 1990’s; have EPA do a study. If that study points out a second round is needed” then “in effect we in the U.S. Congress can vote it up or down.” 136 Cong. Reo. S2739 (daily ed. Mar. 20, 1990), reprinted in 1990 Legislative History, supra, at 5864.

In the meantime, section 202 required EPA to conduct a study of the “need for, and cost effectiveness of, obtaining further reductions in emissions from [ears and certain trucks], taking into consideration alternative means of attaining or maintaining the national ambient air quality standards pursuant to State implementation plans.” CAA § 202(i)(2)(A)(ii), 42 U.S.C. § 7521(i)(2)(A)(ii). EPA must report to Congress on its study no later than June 1,1997. Based on the study, EPA must then determine by December 31, 1999, whether there is a need for new, stricter standards, and whether new standards would be cost effective. CAA §§ 202(i)(2)(B) & (i)(3), 317(a)(5),'42 U.S.C. §§ 7521(i)(2)(B) & (i)(3), 7617(a)(5).

To the extent EPA’s final rule can be viewed as setting emissions standards for new motor vehicles in the northeastern states stricter than those contained in section 202, the rule is inconsistent with the statute.15 *1412EPA nevertheless denies that it violated § 202. States may voluntarily choose to adopt the California standards, pursuant to section 177.16 And so, EPA’s argument goes, it is the northeastern states, not EPA, who will be deciding to impose the California program on their citizens.

There is more than a bit of inconsistency in EPA’s stance. On the one hand, in responding to Virginia’s constitutional arguments, EPA denies that the Ozone Commission has any real power. EPA tells us that it and it alone has the final say under section 184. Any resulting orders of the sort section 184(c)(5) contemplates — orders, in the statute’s words, “requir[ing] each such State to revise its implementation plan to include the approved additional control measures” — are EPA’s, not the Commission’s. But when it comes to sections 202 and 177, EPA asks us to believe that the states who implement the California program in response to EPA’s section 184 order will be doing so voluntarily. Perhaps this is true with respect to states whose governors voted for the Commission’s recommendation, although the legislatures in those states might have something to say on the subject. But no one can suppose that Virginia, whose governor voted against the recommendation, will be freely enacting these restrictive emission standards.

It is time to examine section 177 more closely. It provides that “any State ... may adopt and enforce ... standards relating to control of emissions from new motor vehicles or new motor vehicle engines” if the state standards “are identical to the California standards” and if the standards are adopted “at least two years before” they take effect. CAA § 177, 42 U.S.C. § 7507. As EPA has acknowledged, section 177 gives states the discretion to pass laws containing certain emission standards, laws that would otherwise be preempted. Supplemental Notice, 59 Fed.Reg. at 48,761 col. 1. In other words, a state may decide to follow California’s lead, or it may not.

EPA’s rule takes this choice from the states. To say, as EPA has, that it is “requiring the states to exercise their own independent authority under section 177” is to admit that EPA is depriving the states of that authority. Final Rule, 60 Fed.Reg. at 4718 col. 1. Independence signifies freedom from the dictates of a federal agency. EPA’s rule does not respect the states’ independent authority; it removes it. At several points in its final rule EPA made no bones about this. EPA said its rule would “mandate state action that would otherwise be discretionary,” id. at 4718 col. 2, and that it was “requir[ing] all the northeastern states to adopt the California car program,” id. at 4713 col. 1. And if EPA has chosen those emission standards and mandated their implementation, as it admits it has, the standards are effectively federal standards, in violation of section 177 in light of section 202.

There is another way of approaching this subject. One might treat EPA’s action, not as a requirement that each northeastern state adopt the California standards, but merely as a condition on EPA’s approval of the state’s revised implementation plan. We do not think putting the matter this way changes the outcome. For one thing, the effect of EPA’s rule remains the same despite the difference in formulation. For another, the House Committee Report on section 177 stated: “Finally, the Committee intends these provisions as grants of authority to the States. They are not intended as requirements. Nor may the Administrator require the States to utilize the authorities *1413contained in this section as a condition of approval of a State’s implementation plan.”17 H.R.Rep. No. 95-294, at 311 (1977), reprinted in 1977 U.S.C.C.A.N. 1077, 1390; see also 123 Cong. Reo. 16,675 (1977) (statement of Rep. Rogers) (“No one will force the State to make a judgment. It is left up to the State. They can either do it or not do it.... If a State decides to make that change to clean up the air, clean up the automobile, it can adopt and enforce the California standards which are more strict than the Federal.”); id. at 16,677 (statements of Reps. Wirth and Waxman).

EPA would have us disregard this legislative history because it is pre-1990, when the Act was amended. True enough, the report is old, but so is section 177. The 1990 amendments did not affect the substance of section 177. Congress simply added language preventing the states from prohibiting the sale or manufacture of California-certified vehicles and from taking “any action of any kind to create, or have the effect of creating, a [different type of] motor vehicle.” CAA § 177, 42 U.S.C. § 7507. As explained by Senator Baucus, the author of the Senate bill: “The new language placed in Section 177 ... neither adds to nor detracts from the present authority of the states under Section 177 to adopt, administer and enforce motor vehicle standards that are identical to those adopted by the state of California. The new language simply codifies, in effect, Congress’ understanding of the authority originally” granted “to states by section 177 as expressed in the legislative history of section 177 when it was adopted in 1977.” 136 Cong. Reo. S16,976 (daily ed. Oct. 27, 1990), reprinted in 1990 Legislative HistoRY, supra, at 1021-22.

EPA has another defense against the charge that it has violated section 202 and section 177. The argument is that in enacting section 184 and giving the agency the authority to order states to include control measures in their plans, Congress did not “exclude motor vehicle standards from the measures available to achieve attainment,” Final Brief for Respondent at 41. But this proves much too much. The fact is that section 184 excludes no sort of control measures, of which there are many, some more effective than others. A state might ration gasoline; impose tougher emission standards for boilers, gas turbines, and large internal combustion engines, Supplemental Notice, 59 Fed.Reg. at 48,679 col. 2; encourage carpooling; develop “a comprehensive system of fees and incentives designed to affect driving habits and vehicle usage,” id. at 48,680 col. 2; require “California reformulated gasoline”, id.; enhance “vehicle inspection and maintenance” programs, CAA § 182(c)(3), 42 U.S.C. § 7511a(c)(3); encourage “the voluntary removal from use and the marketplace of pre-1980 model year light duty vehicles and pre-1980 model year light duty trucks,” CAA § 108(f)(l)(A)(xvi), 42 U.S.C. § 7408(f)(l)(A)(xvi).

But among the universe of possible control measures, Congress specifically barred EPA from imposing one particular type — stricter emissions limitations on motor vehicles than those set forth in section 202. And “when a conflict arises between specific and general provisions of the same legislation, the courts should give voice to Congress’s specific articulation of its policies and preferences.” Ohio Power Co. v. FERC, 954 F.2d 779, 784 (D.C.Cir.1992). Here Congress’s policy and preference is loud and clear. It “is the intent of Congress that” EPA not modify the “numerical emission standards” for “any model year before ... 2004.” CAA § 202(b)(1)(C), 42 U.S.C. § 7521(b)(1)(C). EPA therefore may not require, mandate, order, or impose conditions demanding that any state enact particular motor vehicle emission standards, even if those standards are identical to California’s. The time will come when EPA can make its case for tougher emission limitations on motor vehicles. But that time is years from now and, under section 202, the ease must be made to Congress.

*1414Ill

This brings us to the issue we postponed at the beginning of part II of this opinion—whether the record supported EPA’s “SIP caU.” In its final rule EPA declared that each of the state’s implementation plans were “substantially inadequate to comply with the requirements of section 110(a)(2)(D) ... and to mitigate adequately the interstate poUution transport described in section 184 ... to the extent that they do not provide for emissions reductions from new motor vehicles in the amount that would be achieved by the” California.“low emission vehicle ... program.” Final Rule, 60 Fed. Reg. at 4736 col. 2 (to be codified at 40 C.F.R. § 51.120(a)).

EPA’s decision to declare the state plans inadequate depended on its view that it could order the California remedy. Section 184 limits EPA to requiring the control measures the Ozone Commission proposes as “necessary” to cure the problem the Commission identifies. Upon finding the California program “necessary,” EPA became obligated to declare, pursuant to section 110(k)(5), that the states’ plans were inadequate under section 110(a)(2)(D). See CAA § 184(c)(5), 42 U.S.C. § 7511c(e)(5). Section 110(a)(2)(D) provides that state plans must prohibit “emissions activity”—such as driving cars—if this will produce air pollution in amounts that “contribute significantly to nonattainment” in any other state. CAA § 110(a)(2)(D), 42 U.S.C. § 7410(a)(2)(D). EPA’s reasoning was that if the California program is necessary, then the failure of a state to have that program equals a failure (under section 110(a)(2)(D)) to prohibit “significant contribution” to nonattainment in another state. See Final Rule, 60 Fed.Reg. at 4718 col. 3; see also Geoffrey L. Wilcox, New England and the Challenge of Interstate Ozone Pollution Under the Clean Air Act of 1990, 24 B.C. Envtl. Aff. L.Rev. 1, 49 (1996). Tins explains why EPA framed the SIP call as it did—the state implementation plans are inadequate “to the extent they do not provide for reductions from new motor vehicles in the amount that would be achieved” by the California program. 60 Fed.Reg. at 4736 col. 2 (to be codified at 40 C.F.R. § 51.120(a)); see also id. at 4718 col. 3, 4719 col. 1. In short, since the particular measure recommended to EPA is not one the agency can mandate, EPA’s finding of inadequacy—its SIP call— cannot survive.

Furthermore, throughout the rulemaking process, EPA operated on the belief that it had discretion to decide whether to issue a SIP call under section 110, but had no such discretion under section 184. Section 184 forced EPA to review the Ozone Commission’s recommendation, to decide if it was “necessary,” and if it was, to declare all the state plans inadequate. EPA had “no discretion”; it had to find each state plan lacking the measure deficient. But the agency .viewed section 110(k)(5) in a different light. As EPA put it, “if the Administrator were to determine ... that a particular measure is necessary to attain” the National Ambient Standards, “she might exercise her authority under section 110(k)(5) to find that a” State Plan “lacking such measure is substantially inadequate, but she might also refrain from doing so.” Final Brief for Respondent at 64. Thus, although EPA purported to make an independent finding under section 110(k)(5), it is doubtful whether the agency would have exercised the discretion it finds in that provision if section 184 had not been triggered and if the agency had realized that section 110 gave it no power to choose this particular control measure for the states.

Virginia is the only state that has challenged EPA’s rule and we have therefore considered whether to vacate the SIP call only with respect to Virginia. No one has argued for such a ruling. Furthermore, such limited relief would be inconsistent with EPA’s approach and would threaten to further complicate this proceeding. EPA’s modeling was based on an important and, it turns out, invalid assumption: that it could order each northeastern state to adopt the California program. Even if only Virginia’s SIP call were stricken, any findings under section 110 would still have to be revised. If the ozone problem does in fact snowball as it moves up the Eastern seaboard, Virginia’s lack of participation will necessarily affect all areas in the Region downwind of it, namely the rest of the entire Region. The SIP call *1415requires states to “revise” their State Plans “as necessary to correct” them when the “Administrator finds that the applicable implementation plan ... is substantially inadequate to attain or maintain the relevant national ambient air quality standard,” or “to mitigate adequately the interstate pollutant transport described in” sections 176A or 184. CAA § 110(k)(5), 42 U.S.C. § 741000(6). In the absence of applicable modeling, no such agency finding could be made. Thus, there is no existing basis for these SIP calls.

Seven states plus the District of Columbia voted for the California program; only three states (New York, Massachusetts, and Connecticut) have enacted it. It is foreseeable that should the SIP call to other states remain in effect, some will be pressured to adopt the California program. They are, of course, free to do so on their own. But they may not be pushed into that step on the basis of an EPA order that is no longer factually supported. The automobile manufacturers and dealers, who would make and sell different and more expensive cars under the California program, would thus be harmed if we allowed the remaining SIP calls to remain in effect on what is an inadequate record.

Finally, we notice that EPA’s 1996 deadline for areas in moderate nonattainment has passed and that its 1999 deadline for areas in serious nonattainment is drawing near. See CAA § 181(a)(1) Table 1, 42 U.S.C. § 7611(a)(1) Table 1. If EPA considers it appropriate to reexamine state implementation plans in the Northeast, and if EPA requires the states to make revisions in order to bring about further reductions in ozone precursors, it must take these now-passed and rapidly approaching deadlines into account.

IV

In sum, we hold that EPA may not, under section 110, condition approval of a state’s implementation plan on the state’s adoption of a particular control measure. We also hold that EPA may not, under section 184, circumvent section 177, as we interpret it in light of section 202. For the reasons given, we hold that the SIP call EPA issued with respect to each state and the District of Columbia cannot stand.

The petitions for review are granted and the rule is vacated in its entirety.

So ordered.

PER CURIAM:

The Congress enacted the Clean Air Act (the Act), 42 U.S.C. §§ 7401 et seq., “to protect and enhance the quality of the Nation’s air resources so as to promote the public health and welfare and the productive capacity of its population.” Id. § 7401(b)(1). At issue in this case is Title I of the Act, which requires the Environ*145mental Protection Agency (EPA) to promulgate National Ambient Air Quality Standards (NAAQS), thus setting the maximum level of permissible pollutant concentration in the atmosphere. See id. §§ 7408(a)(1), 7409(a)-(b). After the EPA sets the NAAQS, it must determine whether each state is in compliance with these air-quality standards and, in the event of a NAAQS violation, how to establish the geographic boundaries around the non-compliant area. See id. § 7407(d)(1).

In these consolidated petitions, several states, counties, industrial entities and environmental organizations challénge the EPA’s determination that certain geographic areas, are, or are not, in “attainment” with the EPA’s ground-level ozone NAAQS. Id. Some argue that the Act, as applied to them, violates various Constitutional provisions; others argue that the EPA misconstrued the terms of the Act. Virtually every petitioner argues that, for one reason or another, the EPA acted arbitrarily and capriciously in making its final NAAQS designations. But because the EPA complied with the Constitution, reasonably interpreted the Act’s critical terms and wholly satisfied — indeed, in most instances, surpassed — its obligation to engage in reasoned decision-making, we deny the consolidated petitions for review in their entirety.

I. BACKGROUND

The EPA began the odyssey resulting in these consolidated petitions nearly seven years ago. Along the way, it construed a variety of the Act’s provisions, promulgated regulations and issued informal guidance to assist in the collaborative area-designation effort between it and the states. Before discussing the substance of the issues, a brief overview of the Act and the underlying proceedings in this case is in order.

A. The Clean Air Act

Under the Act, the EPA must promulgate NAAQS, which set the maximum ambient, or outdoor, air concentrations for six pollutants that “may reasonably be anticipated to endanger public health or welfare.” 42 U.S.C. § 7408(a)(1). Once it establishes a NAAQS, the EPA must designate each “area” in the United States as “attainment” or “nonattainment.” See id. § 7407(d)(l)(A)(i)-(ii). Alternatively, the EPA may designate an area as “unclassifiable” if the area “permit[s] no determination given existing data.” Catawba Cnty., N.C. v. EPA, 571 F.3d 20, 26 (D.C.Cir.2009) (citing 42 U.S.C. § 7407(d)(l)(A)(i)-(iii)). The EPA treats an “unclassifiable” area as if it were in attainment. See 42 U.S.C. § 7471.

Generally speaking, the EPA designates an area that meets the relevant NAAQS as in attainment, while areas that exceed the NAAQS receive a nonattainment designation. See Catawba Cnty., 571 F.3d at 26. But even if an area’s ambient air concentration complies with the relevant NAAQS, the EPA nonetheless designates it as non-attainment if it “contributes” to a NAAQS violation in a “nearby area.” See 42 U.S.C. § 7407(d)(1)(A)©. The Act does not define the terms “contributes,” “nearby” or “area.”

The EPA works collaboratively with the states to determine the NÁAQS-attainment status for all areas within a respective state’s borders. No later than one year after the EPA promulgates a new or revised NAAQS, each state must submit recommended “initial designations” to the EPA. Id. § 7407(d)(1)(A). A state’s initial designations must suggest both the appropriate geographic boundaries for each “area” and whether the EPA should classify the suggested area as attainment, non-attainment or unclassifiable. See id. § 7407(d)(l)(A)-(B).

*146Once it receives a state’s initial designations, the EPA may either promulgate them as submitted or modify them as it “deems necessary.” Id. § 7407(d)(l)(B)(ii). The Act gives the EPA discretion to change a state’s recommended designation, to alter a state’s proposed geographic area or both. See id. Although the EPA “has no obligation to give any quantum of deference to a designation that it ‘deems necessary’ to change,” Catawba Cnty., 571 F.3d at 40, it must nonetheless notify the state of any intended change and provide the state with at least 120 days “to demonstrate why any proposed modification is inappropriate,” 42 U.S.C. § 7407(d)(l)(B)(ii). These notifications are known as “120-day letters.” See Air Quality Designations for the 2008 Ozone National Ambient Air Quality Standards, 77 Fed.Reg. 30,088, 30,090 (May 21, 2012) [hereinafter 2008 Designations Rule].

While the EPA has ultimate authority to determine each area’s attainment status, each state has “primary responsibility” for ensuring that the geographic areas within its borders either maintain attainment or progress towards it. 42 U.S.C. § 7407(a). Accordingly, once the EPA finalizes its designations, each state must submit to the EPA a State Implementation Plan (SIP) specifying how the NAAQS “will be achieved and maintained.” Id. For areas in attainment, the SIP must simply “contain emission limitations and such other measures as may be necessary ... to prevent significant deterioration of air quality.” Id. § 7471.

For a nonattainment area, however, the Act imposes more stringent requirements. A SIP from a state with a nonattainment area must demonstrate that the state intends to implement “all reasonably available control measures” and “reasonably available control technology” to bring the area into attainment. Id. § 7502(c)(1). The Act also imposes deadlines, or “attainment dates,” on an offending area. See id. § 7502(a)(2)(A). For a violation of a primary1 NAAQS, the offending state must reach attainment “as expeditiously as practicable, but no later than 5 years from the date such area was designated nonattainment.” Id. The EPA “may extend the attainment date to the extent [it] determines appropriate” but only “for a period no greater than 10 years from the date of designation as nonattainment.” Id. Taken together, these two requirements often mean that a state with a nonattainment area must implement potentially expensive technology or expensive process changes to reduce pollution levels over a relatively short period of time. If a state fails to reach attainment timely and the failure is due to inadequate implementation efforts, sanctions can be imposed, including loss of federal highway funds and increasingly severe restrictions on emissions sources within the state. See id. § 7509(a)-(b).

B. The 2008 Ozone NAAQS and the EPA’S 2008 Guidance

On March 12, 2008, the EPA promulgated new primary and secondary NAAQS for ambient ozone,2 a component of urban smog. See 2008 Designations Rule, 77 *147FecLReg. at 30,089. Even though ozone is an “essential presence in the atmosphere’s stratospheric layer,” it becomes harmful at ground level and “can cause lung dysfunction, coughing, wheezing, shortness of breath, nausea, respiratory infection, and in some cases, permanent scarring of the lung tissue.” S. Coast Air Quality Mgmt. Dist. v. EPA, 472 F.3d 882, 887 (D.C.Cir.2006) (quoting Henry A. Waxman, An Overview of the Clean Air Act Amendments of 1990, 21 Envtl. L. 1721, 1758 (1991)). It also “has a broad array of effects on trees, vegetation, and crops and can indirectly affect other ecosystem components such as soil, water, and wildlife.” Mississippi v. EPA, 744 F.3d 1334, 1340 (D.C.Cir.2013). Because ozone forms at ground level when “ozone precursors”— specifically, nitrous oxides (NOx) and volatile organic compounds (VOCs) — react with sunlight, NAAQS compliance largely depends on reducing emissions from ozone-precursor producers like power plants, industrial compounds, motor vehicles and combustion engines. See 2008 Designations Rule, 77 Fed.Reg. at 30,089. Complicating this task is-that ozone and ozone precursors travel easily through the atmosphere, which can result in NAAQS violations hundreds of miles away from the source of the ozone precursors. See id.

Both the EPA’s 2008 primary and secondary ozone NAAQS reduced the maximum allowable daily average eight-hour level of ozone from 0.08 parts per million (ppm) to 0.075 ppm. See National Ambient Air Quality Standards for Ozone, 73 Fed.Reg. 16,436,16,436-37 (Mar. 27, 2008). By setting these new NAAQS, the EPA triggered the states’ responsibility to submit their initial designations. See 42 U.S.C. § 7407(d)(1)(A). To assist this process, the EPA issued a guidance titled “Area Designations for the 2008 Revised Ozone National Ambient Air Quality Standards” [hereinafter 2008 Guidance] on December 4, 2008, which included several matters relevant to the instant petitions.

First, the 2008 Guidance instructed states on the quality of data it expected them to consider. Specifically, it recommended that the states “identify violating areas using the most recent three consecutive years of quality-assured, certified air quality data.” 2008 Guidance at 2. The 2008 Guidance also informed the states that “[i]n general, [NAAQS] violations [will be] identified using data from ... monitors that are sited and operated in accordance with [EPA regulations located at] 40 C.F.R. Part 58.” Id.

Second, the 2008 Guidance provided instruction for establishing geographic boundaries around nonattainment areas, noting first that the “EPA believes it is important to examine ozone-contributing emissions - across a relatively broad geographic area.” 2008 Guidance at 3. Accordingly, the 2008 Guidance recommended that if an air-quality monitor reports a NAAQS violation, the state should consider using the Core Based Statistical Area (CBSA) or Combined Statistical Area (CSA) in which the monitor is located as the “presumptive” boundary.3 Id. If the violating monitor *148is not in a CSA or CBSA, the 2008 Guidance recommended using the county-in which the violating monitor is located as the presumptive boundary. Id.

The 2008 Guidance made plain, however, that CSAs, CBSAs and county lines were merely presumptive boundaries, recognizing that “area-specific analyses ... may support nonattainment area boundaries that are larger or smaller than the presumptive area starting point.” Id. Stressing that “each potential nonattainment area should be evaluated on a case-by-case basis,” the 2008 Guidance instructed the states to consider nine factors when determining a nonattainment area’s borders. See id. at 2, Attach. 2. These include (1) air-quality data; (2) emissions data (such as location of emissions sources and contribution to ozone concentrations); (3) population density and degree of urbanization (including commercial development); (4) traffic and commuting patterns; (5) population growth rates and patterns; (6) meteorology (such as weather and air-transport patterns); (7) geography and topography (such as mountain ranges or other air-basin boundaries that could affect ozone dispersion); (8) jurisdictional boundaries (such as counties, air districts, existing nonattainment area boundaries and regional planning authority boundaries) and (9) the level of control of emissions sources. See id. Attach. 2. The 2008 Guidance stated that the EPA planned to consider these same factors, “along with any other relevant information,” in determining whether to modify the states’ initial designations. Id.

C. The 2008 Ozone Designation Process

By 2009, all states had submitted their initial designations to the EPA. Rather than immediately reviewing the initial designations, however, the EPA halted the designation process to consider whether to lower the ozone NAAQS even further. This delay prompted a lawsuit by Wild-Earth Guardians — an environmental-group petitioner in this case — that sought to compel the EPA to complete the stalled ozone NAAQS designation process.4 The EPA and WildEarth Guardians eventually entered into a consent decree that required the EPA to finalize its designations no later than May 31, 2012. See 2008 Designations Rule, 77 Fed.Reg. at 30,091.

The EPA notified the states in September 2011 that it intended to finalize the ozone NAAQS designations by the May 31, 2012 deadline set forth in the consent decree. In accordance with the 2008 Guidance’s instruction to “identify violating areas using the most recent three consecutive years of quality-assured, certified air quality data,” 2008 Guidance at 2, virtually every state had already submitted air-quality data from 2008 to 2010 by the time the EPA resumed the designation process. Although the EPA assured the states that it still planned to consider the recommended designations and ozone data they had submitted initially, it recognized that some states may have collected more recent air-quality data for their regions. For this reason, the EPA allowed the states to provide updated recommendations and analyses — so long as any updated air-quality data was certified for quality — but assured them that they were under no obligation to do so. In response to this invitation, several states updated their initial designations and some submitted air-quality data from 2009 to 2011 to replace their older 2008 to 2010 data. The states seeking to use data from 2009 to 2011 agreed to certify their data for *149quality by February 29, 2012, so that the EPA had sufficient time to consider the more recent data in advance of its May 31, 2012 deadline to finalize the designations.

The EPA then reviewed each state’s initial designations to determine whether to modify them. It first examined the air-quality submissions from the states to determine which monitors reported ozone NAAQS violations. If a state certified its air-quality data from 2011 by the February 29, 2012 deadline, the EPA generally considered its air-quality data from the years 2009 to 2011. For all other states, the EPA considered air-quality data from 2008 to 2010.

After identifying NAAQS-violating monitors, the EPA decided whether to alter the states’ respective recommended nonat-tainment boundaries. To do so, the EPA used a multi-factor, weight-of-the-evidence test that tracked — but was not identical to — the nine-factor test in the 2008 Guidance. Specifically, the EPA collapsed the 2008 Guidance’s nine-factor test into a five-factor test, which examined (1) “Air Quality Data,” or whether an area’s monitor reported a NAAQS violation; (2) “Emissions Data,” including emissions levels and controls, population, population density, population growth, degree of urbanization and traffic and commuting patterns; (3) “Meteorology,” including wind speed and direction; (4) “Geography/Topography,” which examined the effect of physical land features on the distribution of ozone and (5) “Jurisdictional Boundaries,” which helped determine whether certain areas could effectively carry out air-quality planning and enforcement functions for nonat-tainment areas.

Once attainment designations were made, the EPA notified the states of any proposed modifications it deemed necessary and invited them to submit any additional data or comments they wished to have the EPA consider. Although not required by statute, see 42 U.S.C. § 7407(d)(2)(B), the EPA also opened a 30-day public comment period on the proposed notifications. Several states, organizations and members of the public — including many of the petitioners in this case— submitted comments. The EPA considered the comments and then promulgated its final designations, which identified 48 nonattainment areas in 26 states, the District of Columbia and Indian country. The nonattainment areas included 192 counties in toto and 36 counties in part. The EPA published the majority of its final designations on May 21, 2012, see 2008 Designations Rule, 77 Fed.Reg. at 30,088, and in the case of certain Chicago-area designations, on June 11, 2012, see Air Quality Designations for the 2008 Ozone National Ambient Air Quality Standards for Several Counties in Illinois, Indiana, and Wisconsin; Corrections to Inadvertent Errors in Prior Designations, 77 Fed.Reg. 34,221, 34,221 (June 11, 2012).

After the EPA received and denied 29 petitions for reconsideration, the parties in this consolidated case5 petitioned this *150Court for review. We have jurisdiction under 42 U.S.C. § 7607(b)(1).

II. COMMON LEGAL PRINCIPLES

Before addressing the petitioners’ individual challenges, we think it helpful to discuss several principles that bear on most, if not all, of the issues the petitioners have raised.

First, we review the EPA’s NAAQS designations under the same standard we use in reviewing a challenge brought under the Administrative Procedure Act (APA). See Allied Local & Reg’l Mfrs. Caucus v. EPA 215 F.3d 61, 68 (D.C.Cir.2000). Accordingly, we will set aside a NAAQS designation by the EPA only if it is “arbitrary, capricious, an abuse of discretion, or otherwise not in accordance with law.” Catawba Cnty., 571 F.3d at 41 (quoting 5 U.S.C. § 706(2)(A)). We must, however, give an “extreme degree of deference” to the EPA’s evaluation of “scientific data within its technical expertise,” City of Waukesha v. EPA, 320 F.3d 228, 247 (D.C.Cir.2003), especially where, as here, we review the “EPA’s administration of the complicated provisions of the Clean Air Act.” Catawba Cnty., 571 F.3d at 41 (citing Nat’l Ass’n of Clean Air Agencies v. EPA 489 F.3d 1221, 1229 (D.C.Cir.2007)). Because the EPA’s “basic obligation” is to conduct “reasoned decisionmaking,” id. at 25, we will uphold its action if the record shows that the EPA “considered all relevant factors and articulated a ‘rational connection between the facts found and the choice made,’ ” id. at 41 (quoting Burlington Truck Lines, Inc. v. United States, 371 U.S. 156, 168, 83 S.Ct. 239, 9 L.Ed.2d 207 (1962)).

Second, we have long since rejected the argument that the EPA violates the Act if it uses a holistic, multi-factor, weight-of-the-evidence test for determining whether a given area contributes to a NAAQS violation. See ATK Launch Sys., Inc. v. EPA, 669 F.3d 330, 336-37 (D.C.Cir.2012) (challenge to 2006 fine particulate matter NAAQS designations); Catawba Cnty., 571 F.3d at 46 (challenge to 1997 fine particulate matter NAAQS designations). Indeed, in Catawba County, we made explicit that the EPA does not violate the Act even if it fails to adopt “a bright-line, ‘objective’ test” for determining contribution and we also held that the “EPA’s failure to quantify its analysis” does not render “its interpretation of ‘contribute’ arbitrary and capricious and therefore unreasonable.” 571 F.3d at 39. Rather, because “[a]n agency is free to adopt a totality-of-the-circumstances test to implement a statute that confers broad discretionary authority, even if that test lacks a definite ‘threshold’ or ‘clear line of demarcation to define an open-ended term,’” we have held that, “[t]o be reasonable, such an ‘all-things-considered standard’ must simply define and explain the criteria the agency is applying.” Id.

With this background in mind, we now turn to the petitioners’ challenges.

III. THE PETITIONERS’ CHALLENGES

A. Delaware & Connecticut

We begin with a challenge to the EPA’s construction of the key statutory provision in this case. Petitioners Delaware and Connecticut challenge the EPA’s refusal to designate broad, multi-state non-attainment areas to address the issue of long-range ozone transport. According to the States, the EPA’s final designations are inconsistent with its statutory mandate to designate areas as nonattainment if they “contributef ] to ambient air quality in a nearby area that does not meet [the NAAQS].” 42 U.S.C. § 7407(d) (emphasis added). We conclude, to the contrary, that the designations are consistent with the EPA’s reasonable interpretation of the ambiguous statutory term “nearby.”

*151After the EPA reopened the designation process in 2011, Delaware proposed a non-attainment area that would stretch across 16 upwind states and the District of Columbia — to states as far west as Missouri. Connecticut similarly proposed an 18-state nonattainment area, also stretching west to Missouri. Both States argued for what Delaware described as a “more workable definition of ‘nearby ” — one that would ask “whether a source is ‘near enough to contribute’ to nonattainment or interfere with maintenance.” Letter from Del. Dep’t of Natural Res. & Envtl. Control to EPA 5 (Oct. 28, 2011) [hereinafter Delaware Response].

The EPA, however, had taken a different approach in the 2008 Guidance, instead interpreting “nearby” as presumptively including counties in the same metropolitan area as the violating county. 2008 Guidance at 3. In the Guidance, the EPA acknowledged that certain regions have ozone transport problems, but it concluded that the Act “does not require that all contributing areas be designated nonat-tainment, only the nearby areas.” Id. at 4. The agency explained that “[regional strategies, such as those employed in the Ozone Transport Region and EPA’s NOx SIP Call are needed to address the long-range transport component of ozone nonat-tainment.” Id. In keeping with this understanding of the statute, the EPA declined to designate “super-regional” nonattainment areas, see Responses to Significant Comments on the State and Tribal Designation Recommendations for the 2008 Ozone NAAQS at 8-9 (Apr. 30, 2012) [hereinafter Response to Comments], and instead made more limited nonattainment designations in both Delaware and Connecticut, see Delaware Area Designations for the 2008 Ozone NAAQS 2; Connecticut Area Designations for the 2008 Ozone NAAQS l.6

We evaluate the EPA’s interpretation of a Clean Air Act provision under the familiar two-step Chevron framework. See Util. Air Regulatory Grp. v. EPA, — U.S. —, 134 S.Ct. 2427, 2439, 189 L.Ed.2d 372 (2014) (citing Chevron, U.S.A., Inc. v. Natural Res. Def. Council, Inc., 467 U.S. 837, 842-43, 104 S.Ct. 2778, 81 L.Ed.2d 694 (1984)). The first question— “whether Congress has directly spoken to the precise question at issue,” Chevron, 467 U.S. at 842, 104 S.Ct. 2778 — has previously been resolved by this Court. In Pennsylvania Department of Environmental Protection v. EPA (PADEP), we held that the statutory term “nearby” in section 107(d) is ambiguous; indeed, we reached that conclusion in the course of addressing the precise argument that Delaware makes here. See 429 F.3d 1125, 1129-30 (D.C.Cir.2005). In Catawba County, we reached the same conclusion. See 571 F.3d at 35 (noting that section 107(d) does not define “nearby,” and that it is “the kind[ ] of word[ ] that suggests] a congressional intent to leave unanswered questions to an agency’s discretion and expertise”).

Recognizing these precedents, Delaware and Connecticut conceded at oral argument that our analysis must be governed by Chevron’s second step, Oral Arg. Recording at 3:49-3:54, which requires us to ask only whether the EPA’s interpretation is reasonable, see, e.g., PADEP, 429 F.3d at 1130; But we have addressed that question once as well, also in PADEP, where we said that “Chevron requires that we defer to the agency’s reasonable interpretation of the term, and Delaware has given us no reason to think that EPA’s interpretation is unreasonable.” Id. We reach the same conclusion here.

*152First, the agency’s interpretation of “nearby” — as presumptively including counties within the same metropolitan area as the violating county — falls readily within the dictionary definition of “nearby” as “close at hand; not far off; adjacent; neighboring.” RANDOM House College DICTIONARY 889 (rev. ed.1980). By contrast, neither the dictionary nor common parlance would regard Missouri as “nearby” to Connecticut or Delaware, as the petitioners’ proposals would require.

Second, the EPA’s construction is consistent with the approach the agency has taken in prior designations proceedings— an approach that this Court has previously upheld as reasonable. See PADEP, 429 F.3d at 1127, 1129-30; 2008 Guidance at 3.

Third, the EPA’s construction is consistent with the statutory scheme. The EPA selected the metropolitan area as the presumptive “nearby” area for its contribution analysis in part because the Congress itself chose the metropolitan area as the default boundary for ozone nonattainment areas classified as “serious,” “severe,” or “extreme.” ' See 42 U.S.C. § 7407(d)(4)(A)(iv); 2008 Guidance at 3 n. 5. The Congress’ choice is certainly evidence that the legislature envisioned broad but relatively local nonattainment areas.7

As in PADEP, the petitioners argue that the EPA’s interpretation.is unreasonable because it fails to appreciate the role of ozone transport, and consequently yields designations that fail to include the true contributors to their nonattainment status. See PADEP, 429 F.3d at 1129-30. Delaware notes, for example, that 84 to 94 per cent of its ozone results from the contributions of other states, including states as far west as Missouri. See Delaware Reply Br. 4. Without emissions reductions from those states, petitioners argue,- they cannot meet the 0.075 ppm standard. Thus, by failing to address the principal sources of their ozone pollution, the EPA’s interpretation eliminates any possibility that they will attain the NAAQS.8

Athough we are sympathetic to the petitioners’ concerns, our role is not to decide whether their proposed interpretation is reasonable. Instead, the sole question before us is whether the EPA interpreted the term reasonably and consistently with the statute. See PADEP, 429 F.3d at 1130 (noting that, although a broader “construction of ‘nearby’ may well be sensible, Chevron requires that we defer to the agency’s reasonable interpretation of the term”). Here, the EPA had already considered the problem thé petitioners raised. Part of the rationale for using the metropolitan area as the starting point for the contribution analysis was to account for ozone transported from outside the violating county. See -2008 Guidance at 3--A. A-though this approach does not fully account for longer-range, interstate transport, the EPA has addressed that problem in regulations promulgated under other provisions of the Act. See, e.g., Federal Implementation Plans: Interstate Transport of Fine Particulate Matter and Ozone *153and Correction of SIP Approvals, 76 Fed. Reg. 48,208 (Aug. 8, 2011) (promulgating the Cross State Air Pollution Rule, commonly referred to as the Transport Rule).9 Although the petitioners recognize the EPA’s reliance on those other regulatory options, they maintain that they “have been less than successful” up to this point. Delaware Br. 6; see also id. at 9. We, however, must defer to the EPA’s reasonable judgment that regional strategies adopted pursuant to other statutory provisions specific to long-range ozone transport remain the appropriate means for addressing this problem. See 2008 Guidance at 4.

The petitioners note that our decision in PADEP rested in part upon the fact that there, Delaware had “offered no evidence that ‘in practice’ EPA will not enlarge a nonattainment area in response to [its then] eleven-factor analysis.” 429 F.3d at 1130. Indeed, in PADEP, Delaware had failed altogether “to produce an eleven-factor analysis.” Id. But we did not mean by this to suggest that, had Delaware produced the appropriate factor analysis, the EPA would have been required to adopt an interpretation of “nearby” that included states as far away as those within the-petitioners’ proposed nonattainment areas. The points discussed above — including the dictionary definition of “nearby” and the consistency of the EPA’s interpretation with the statute and its prior practice— strongly suggest that the EPA’s narrower interpretation would still be reasonable.

Nonetheless, if the petitioners had submitted a persuasive five-factor analysis establishing contributions from farther-away states, that would be relevant to our assessment of the reasonableness of the EPA’s refusal to enlarge the nonat-tainment area beyond its presumptive scope. In this case, however, although the petitioning States did submit technical analyses, they failed to demonstrate the requisite linkages under the EPA’s 2008 Guidance. See, e.g., Delaware Response Attach. 2 at 5-7, 11-13 (disputing relevance of factors related to urbanization, traffic, and economic growth); id. at 14-15 (with respect to meteorology factor, describing long-range transport without describing weather patterns within the proposed 16-state nonattainment area). Hence, the petitioners did not show that the agency “will not enlarge a nonattainment area in response to” the (current) five-factor analysis, PADEP, 429 F.3d at 1130. Rather, the States’ analyses were simply insufficient to overcome the agency’s definitional presumption.

In sum, we conclude that the EPA’s final designations of Delaware and Connecticut counties are consistent with a reasonable interpretation of the Clean Air Act.10

*154B. Uinta Basin

Petitioner WildEarth Guardians (Wild-Earth) challenges the EPA’s designation of Uinta Basin, Utah, as “unclassifiable.” We find the EPA’s designation rational and in accordance with the Clean Air Act, and we therefore deny WildEarth’s petition.

1. Uinta Basin Background

The EPA requires every state to establish a network of regulatory monitoring stations to collect ozone air-quality data. See 40 C.F.R. pt. 58. The number of regulatory monitors required in an area depends, in part, on the area’s population. See id. app. D. tbl.D-2. Areas with populations below 50,000 and many areas with fewer than 350,000 inhabitants require no regulatory monitors. Id. Many rural areas therefore lack monitors.

Uinta Basin, Utah, had no regulatory monitoring until April 2011. The pre-2011 absence of regulatory-air-quality monitors in Uinta Basin meant that, when the EPA in 2013 conducted the designation process for the 2008 NAAQS, the agency had regulatory data for Uinta Basin for only two years — 2011 and 2012. The 2008 ozone NAAQS, however, reflect three-year averages of ozone levels. See 2008 Designations Rule, 77 Fed.Reg. at 30,089. Noting that “there are not yet three consecutive years of certified ozone monitoring data available [from Uinta Basin] that can be used to determine the area’s attainment status,” id., the EPA designated the area as “unclassifiable,” which the Clean Air Act defines as an area that “cannot be classified on the basis of available information as meeting or not meeting” the NAAQS, 42 U.S.C. § 7407(d)(l)(A)(iii).

Although no regulatory data exist for Uinta Basin prior to 2011, private companies working under consent decrees have been required to operate ozone air-quality monitors in Uinta Basin since 2009. See Letter from Robin Cooley, Counsel, Wild-Earth Guardians to Lisa P. Jackson, Adm’r, EPA 3 (July 19, 2012). Under the terms of those consent decrees, the private monitors must comply with many of the same requirements as regulatory monitors. See Consent Decree ¶¶ 80-81, United States v. Kerr-McGee Corp., No. 1:07-cv-01034 (D. Colo. May 17, 2007). From 2009 to 2011, the private monitors provided raw data showing ozone levels significantly exceeding the 2008 ozone NAAQS. The EPA found the 2009 to 2011 private data insufficient to support a nonattainment designation.

2. The Private Monitoring Data Challenge

WildEarth argues that, inflight of the private data, the EPA contravened the Act’s requirements when it designated Uinta Basin as unclassifiable rather than nonattainment. We disagree.

The Act calls for the EPA to make designations “on the basis of available information.” 42 U.S.C. § 7407(d)(l)(iii). We have repeatedly found similar language to be ambiguous when assessing whether to defer to an agency’s construction. See Catawba Cnty., 571 F.3d at 35, 38 (finding the phrase “based on air quality monitoring data” to be ambiguous); Sierra Club v. EPA, 356 F.3d 296, 305-06 (D.C.Cir.2004) (finding the phrase “based on photochemical grid modeling” to be ambiguous). The EPA therefore may interpret the statutory language as it sees fit, as long as its interpretation is reasonable. Chevron, 467 U.S. at 845, 104 S.Ct. 2778. And even assuming the Act obligates the EPA to consider certain types of data, there would be no obligation for the agency to base its designations on data it reasonably considers to be unsound, at least if it “adequately explain[s] its reasons for rejecting ... data” on which it declines to *155rely. City of Waukesha, 320 F.3d at 248. We evaluate the EPA’s reasons cognizant of the “extreme degree of deference” we owe an agency “when it is evaluating scientific data within its technical expertise.” Catawba Cnty., 571 F.3d at 41.

The EPA reasonably explained that the private monitoring data afforded an insufficient basis for a nonattainment designation because the agency was unable to perform post-collection quality assurance checks on the data. In particular, the EPA lacked quality assurance data needed to verify and audit the private data. As the agency explained:

Quality assurance data consist, primarily, of biweekly single point quality control (QC) checks, used to assess the precision and bias a given instrument is displaying in its day-to-day measurements, and annual independent performance evaluations (audits) of equipment, which rely on independent staff and measuring systems to confirm that the monitors are operating as expected and required.

Letter from Lisa P. Jackson, Adm’r, EPA to Robin Cooley, Counsel, WildEarth Guardians 5 (Dec. 14, 2012) (denying reconsideration of Uinta Basin designation). The agency determined that, without audits or quality control checks, it could not adequately verify the quality of the private data. That explanation comports with common sense and falls within the substantial deference accorded the EPA in evaluating the soundness of data available to it.

WildEarth presses several counterarguments, none of which we find persuasive. First, WildEarth observes that the consent decrees required the private monitors to operate in “substantial compliance” with 40 C.F.R. Part 58, the quality assurance requirements under which regulatory monitors operate. But “substantial compliance” is not “full compliance,” and the EPA could reasonably draw a distinction between the two. Moreover, data from regulatory monitors — which must be collected in compliance with 40 C.F.R. Part 58 — undergo post-collection auditing and verification processes. See, e.g., .40 C.F.R. pt. 58, app. A, § 3. Those post-collection processes could not be conducted for the private monitor data. Accepting Wild-Earth’s argument would require us to conclude that the EPA must apply less stringent post-collection validation requirements to data collected from private monitors in “substantial compliance” with the agency’s data-collection regulations than the agency applies to data collected from régulatory monitors in actual compliance with those regulations. We see no reason to embrace that counterintuitive result.

Second, WildEarth points out that the EPA has encouraged other federal entities to take notice of the private monitoring data. The EPA acknowledges that it argued, in a judicial proceeding supporting entry of the same consent decrees mandating the private monitoring, that the private monitors would provide data that would be “reliable and of good quality” and “useful in assisting regulators.” Resp’t’s Br. 57. And indeed the data have proven helpful to the EPA in other regulatory contexts. On the basis of the private data, for example, the EPA informed the Forest Service that Uinta Basin ozone concentrations “exceed the NAAQS” and are a “serious problem.” Supp. JA 387.

We agree with WildEarth that an agency may be required to articulate why data are sufficiently reliable for one purpose but not for another. See Cnty. of L.A. v. Shalala, 192 F.3d 1005, 1022 (D.C.Cir.1999). But the EPA has done so here. That the data may be sufficiently reliable to warrant identifying ozone as a serious issue for a Forest Service analysis under *156one statutory provision does not necessarily mean that the data are reliable enough to compel a nonattainment designation under a different statutory regime. To hold otherwise would require the EPA wholly to blind itself to potentially useful private data for any purpose if it were to consider that data insufficiently reliable for one purpose. There is no basis for constraining the agency in that way.

That the EPA partially relied on the private data in the course of this very designation process does not undercut that conclusion. While “unclassifiable” represents a single statutory designation, see 42 U.S.C. § 7407(d)(l)(A)(i)-(iii), the EPA further divided that classification into two sub-categories: “unclassifiable/attainment” and “unclassifiable.” See 2008 Designations Rule, 77 Fed.Reg. at 30,089. “Historically for ozone,” the EPA designates as “ ‘unclassifiable/attainment’ ” those areas for which “air quality information is not available because the areas are not monitored.” Id. at 30,090. But in Uinta Basin, the EPA instead designated the area “unclassifiable” after determining that the private monitoring “detected levels of ozone that exceed the NAAQS.” Id. at 30,089.

There is no arbitrariness in the EPA’s choice partially — but not fully — to rely on the private data. At the outset, we note that the parties point us to no material differences between an “unclassifiable/attainment” and an “unclassifiable” designation, and we are aware of none. See 40 C.F.R. § 51.1100(g) (“Attainment area means, unless otherwise indicated, an area designated as either attainment, unclassifiable, or attainment/unclassifiable.”); cf. 42 U.S.C. § 7471 (instructing the EPA to give the same treatment to “unclassifiable” and “attainment” areas for SIP purposes). But give'n the EPA’s decision to create two different unclassifiable designations, we will assume arguendo that materially different regulatory burdens attend each designation. Even then, however, we agree with the EPA that it was reasonable to conclude that it would be inappropriate to label the Uinta Basin area “unclassifiable/attainment”: the private data, even if unverified, at least implied that a NAAQS violation was possible, even if not conclusively proven to the agency’s satisfaction. WildEarth, moreover, points to no other area for which private — but not regulatory — monitoring suggested a NAAQS violation. It thus appears that Uinta Basin differed from all other areas meriting an “unclassifiable/attainment” designation. We conclude that the EPA’s conclusion partially — but not fully — to credit the private data was reasonable and non-arbitrary, particularly in light of the “extreme deference” we owe the agency. See Catawba Cnty., 571 F.3d at 41.

In sum, the EPA reasonably declined to rely on data that it considered of insufficient quality for designations purposes. With that conclusion, and having reviewed the remainder of WildEarth’s challenges and determined that they lack merit, we deny the group’s petition for review. See Catawba Cnty., 571 F.3d at 52.

C. Sierea Club

Petitioner Sierra Club challenges the EPA’s refusal to use uncertified 2011 air-quality data during the designation process, a decision that resulted in 15 counties avoiding nonattainment designations. Finding the EPA’s actions rational and in accordance with the Clean Air Act, we deny Sierra Club’s petition.

1. Sierra Club Background

In furtherance of the Clean Air Act’s “ ‘core principle’ of cooperative federalism,” EPA v. EME Homer City Generation, L.P., — U.S. —, 134 S.Ct. 1584, 1602 n. 14, 188 L.Ed.2d 775 (2014), states take the lead in the collection of air-quality *157data. In doing so, states operate regulatory monitors under an array of “[exhaustive technical specifications” promulgated by the EPA. Catawba Cnty., 571 F.3d at 30; see 40 C.F.R. pt. 58. States “edit[]” and “validate[ ]” the collected data pursuant to the EPA-mandated procedures and report it to the EPA according to a prescribed schedule. See 40 C.F.R. .§ 58.16(b)-(c). Data collected in each quarter must be “edited, validated and entered” into the EPA’s system within ninety days of the end of the quarter. Id. “For example, the data for the reporting period January 1 — March 31 are due on or before June 30 of that year.” Id. § 58.16(b). Post-auditing, the data are still considered “uncertified” when submitted to the EPA.

While uncertified data from the first quarter (ie., January 1 to March 31) become available to the EPA as of June 30, those data remain subject to continuing audits and edits by states. The data collection process reaches completion only when a state provides final certification that the necessary “ambient concentration and quality assurance data are completely submitted ... and ... are accurate.” Id. § 58.15(a). The EPA requires certification by May 1 of the following calendar year for all data collected in the previous year. Id. § 58.15(a)(2). States therefore had to certify their 2011 data by May 1, 2012.

As explained, because the 2008 ozone NAAQS represent a three-year average, the EPA needs air-quality data from three sequential calendar years to classify an area as attainment or nonattainment (as opposed to unclassifiable). See 2008 Designations Rule, 77 Fed.Reg. at 30,089. In the designation process for the 2008 NAAQS, the EPA gave each state a choice between two options: (i) early-eertify 2011 data by February 29, 2012, in which event the EPA would consider 2009 to 2011 data for the designation process for that state (Option One); or (ii) decline to early-certify (and stick to the normal May 1 certification deadline), in which event the EPA would use 2008 to 2010 data for designations in that state (Option Two). See id. at 30,091.

At least eight states selected Option Two. Sierra Club identifies over one dozen counties within those eight states for which the choice between Option One and Option Two ii.e., the choice between designations based on 2008 to 2010 data versus 2009 to 2011 data) allegedly meant that those counties avoided nonattainment designations. See Letter from Robert Ukei-ley, Counsel, Sierra Club to EPA, Re: Designations for the 2008 Ozone NAAQS Docket ID No. EPA-HQ-OAR-2008-0476 at 3 tbl.l (Feb. 3, 2012). Sierra Club contends that the EPA was compelled to use 2009 to 20Í1 data for those areas. We disagree and conclude that the EPA’s actions were non-arbitrary.

2. Uncertified Data Challenge

Sierra Club first notes that, at the time of the designation process, the EPA possessed uncertified 2011 data for all areas. Because the agency’s regulations require the submission of uncertified data within ninety days of the end of the quarterly reporting period, see 40 C.F.R. § 58.16(b), the EPA had all 2011 uncertified data in its possession by the end of March. It should have used that data, Sierra Club argues, notwithstanding the lack of certification. We are unpersuaded.

While the uncertified data must undergo preliminary auditing and quality checks before submission to the' EPA, see id. § 58.16(c), those preliminary quality control measures are just that — preliminary. As the EPA explains, the data remain subject to continuing checks and revisions by the states until final certification. Resp’t’s Br. 66. Accordingly, the EPA *158reasonably “does not presume that data [validation and auditing] processes are complete and accurate until” the final data certification. Id. at 46. Mindful of the significant deference we owe the EPA in matters concerning data quality or sufficiency, see Catawba Cnty., 571 F.3d at 41, we see no basis for second-guessing the EPA’s considered judgment on the issue.

Sierra Club next argues that, even if the agency acted reasonably in refusing to rely on uncertified data, it acted arbitrarily in declining to delay the designation process until all states had certified their 2011 data by the standard May 1 deadline. After all, Sierra Club notes, the consent decree under which the EPA. conducted the designation process allowed the agency until May 31, 2012, to promulgate the final designations. 2008 Designations Rule, 77 Fed. Reg. at 30,091.

Sierra Club, however, identifies no authority obligating the EPA to wait until the last possible minute to promulgate its designations. And in this case, doing so would have made little sense. The EPA entered into the consent decree precisely to settle allegations that it had already missed the Act’s statutory deadlines for promulgating the 2008 ozone NAAQS designations. See id. Accepting Sierra Club’s position would effectively call for the EPA to infringe the Act’s deadlines still further. In any event, as the EPA explained in denying Sierra Club’s petition for reconsideration of the designations after the May 1, 2012, certification deadline passed and 2009 to 2011 data were fully certified and available to the EPA, “[n]ew technical data become available on a regular basis.” Letter from Lisa P. Jackson, Adm’r, EPA to Robert Ukeiley, Counsel, Sierra Club enclosure p. 2 (Dec. 14, 2012). The EPA reasonably concluded that delay “to consider such new information would result in a never-ending process in which designations are never finalized.” Id. Indeed, Sierra Club itself has already filed a petition for reconsideration based on 2010 to 2012 data. See Sierra Club Reply Br. 8. The EPA could reasonably conclude that the process must end at some point. We conclude that the agency did not act arbitrarily in ending it here. Cf. Catawba Cnty., 571 F.3d at 51 (“New York’s underlying complaint is that the iterations should have continued, perhaps ad infini-tum. But such a process is inconsistent with the CAA: Congress imposed deadlines on EPA and thus clearly envisioned an end to the designations process.”).

With that conclusion, and having reviewed the remainder of • Sierra Club’s challenges and determined that they lack merit, we deny the group’s petition for review. See Catawba Cnty., 571 F.3d at 52.

D. Mississippi

The State of Mississippi challenges the EPA’s use of 2008 to 2010 data to classify the counties within the Memphis, Tennessee area, an analysis that resulted in a nonattainment designation for part of De-Soto County, Mississippi. Because we conclude that the EPA’s actions were rational and in accordance with the Clean Air Act, we deny Mississippi’s petition for review.

1. Mississippi Background

In Mississippi and elsewhere, the EPA conducted the designations for metropolitan areas through a two-step process. First, the EPA examined air-quality data from all regulatory monitors in a metropolitan area. If no monitors in the area showed a NAAQS violation, no county in the area would be designated nonattainment. In that event, there would be no second step. But if a single monitor from the area showed a NAAQS violation, the county housing the violating monitor would *159be designated nonattainment. See 2008 Guidance at 3-4. In that case, the EPA would proceed to the second step for that metropolitan area.

The second step took account of the fact that the Act mandates nonattainment designations not only for areas themselves exceeding the relevant NAAQS, but also for all areas that “contribute[ ]” to a NAAQS violation in a “nearby area,” even if the “contributing” area’s air quality' — ■ considered alone — meets the NAAQS. See 42 U.S.C. § 7407(d)(l)(A)(i); 2008 Guidance at 3-4. In the second step, the EPA assessed each county in a metropolitan area with a violating monitor on a case-by-case basis to determine if the county contributed to the identified violation. If, on the basis of a multi-factor test, the EPA determined that a county “contributed” to the NAAQS exceedance at the violating monitor in another county, the EPA also designated the contributing county as non-attainment. We have repeatedly upheld multi-factor contribution analyses as consistent with the Act’s designation process under section 107 — a conclusion that Mississippi does not challenge here. See, e.g., ATK Launch Sys., 669 F.3d 330; Catawba Cnty., 671 F.3d 20. See generally supra § II.

In 2011 and 2012, the EPA conducted that two-step designation process for the Memphis CBSA. The Memphis CBSA consists of several counties in Tennessee (Shelby, Tipton, and Fayette), Mississippi (DeSoto, Marshall, Tate, and Tunica), and Arkansas (Crittenden). See Office of Mgmt. & Budget, OMB Bulletin No. 10-02, Update of Statistical Area Definitions and Guidance on Their Uses 40 (Dec. 1, 2009). At the first step, the EPA evaluated 2008 to 2010 certified air-quality data and detected a NAAQS violation at the monitor in Shelby County, Tennessee. Proceeding to the second step, the EPA conducted the multi-factor analysis and determined that part of DeSoto County, Mississippi, contributed to the Shelby County violation.

On December 9, 2011, the EPA notified Mississippi that it planned to designate part of DeSoto County as nonattainment when it promulgated the final designations in 2012. The EPA invited Mississippi (and all other states) to provide to the agency by February 29, 2012, any additional information for consideration in the final designation process — including any early-certified 2011 data. See Memphis, TN-MS-AR Area Designations for the 2008 Ozone NAAQS 3-4 [hereinafter Memphis Area Designations]. Mississippi responded to the EPA’s multi-factor analysis with its own multi-factor analysis, disputing the EPA’s conclusion that DeSoto County contributed to any violation in Shelby County. Additionally, Mississippi and Tennessee— two of the three states in the Memphis CBSA — early-certified their 2011 data before the February 29, 2012, deadline. Arkansas — the third state in the Memphis CBSA — declined to early-certify any 2011 data.

On May 21, 2012, the EPA published its final designations for the Memphis CBSA. At the first step of the two-step designation pro'cess, the agency used 2008 to 2010 data and again identified a violation at the Shelby County monitor. The EPA then moved to the second step and, after considering Mississippi’s multi-factor analysis and updating its own analysis accordingly, reiterated its original conclusion that part of DeSoto County contributed to the Shelby County violation. The agency therefore designated part of DeSoto County as nonattainment. See Memphis Area Designations at 16. Mississippi claims that designation was arbitrary and capricious. We disagree.

*1602. Challenge to the First Step of the Designation Process

First, Mississippi argues that the EPA acted arbitrarily in using 2008 to 2010 data for the first step of the two-step designation process (ie., identifying violating monitors within a CBSA) even though the EPA possessed early-certified 2011 data from Tennessee. The 2009 to 2011 data showed no NAAQS violation at the Shelby County monitor. Accordingly, Mississippi argues, no violation should have been identified at the first step of the two-step designation process. But the EPA declined to evaluate Shelby County using the early-certified 2009 to 2011 data, instead using the 2008 to 2010 data. True, the EPA must adequately explain why it declined to rely on the early-certified 2011 data. See City of Waukesha, 320 F.3d at 248. But the agency did so.

At the time of the final designations, the EPA had in its possession early-certified data from Mississippi and Tennessee, but not from Arkansas. In the first step of its two-step designation process, the EPA evaluates all air-quality monitors in a metropolitan area. Without 2011 Arkansas data, the EPA did not have a full set of 2011 data for the Memphis CBSA. The EPA only had data from different time horizons — 2008 to 2010 data for the Arkansas portion of the Memphis CBSA, and 2009 to 2011 data for the Tennessee and Mississippi portions of that same CBSA. The agency declined to rely on this mismatched dataset. Instead, the EPA opted to rely on the most recent matched dataset in its posséssion: the complete set of 2008 to 2010 data. We see no reason — and Mississippi provides none — to declare irrational the EPA’s conclusion that comparing data from the same time period would be more appropriate than analyzing data from different time periods in the same evaluation process. Cognizant of the substantial deference we owe the EPA in that highly technical evaluation, see Catawba Cnty., 571 F.3d at 41, we find the EPA was entitled to rely on a matched dataset instead of a mismatched one.

Even assuming the EPA’s choice to rely only on matched datasets for the Memphis CBSA was reasonable (as we conclude it to be), Mississippi argues that the EPA’s approach nonetheless was arbitrary because the agency required a matched dataset for Memphis-area designations but allegedly relied on a mismatched dataset for Chiea-go-area designations. “[I]nconsistent treatment,” we have found, is a “hallmark of arbitrary agency action.” Id. at 51. There was no inconsistent treatment here, however. In both Chicago and Memphis, the EPA relied only on matched datasets in the designation process.

With regard to the Chicago metropolitan area, Illinois early-certified its 2011 data. Wisconsin and Indiana — portions of which also lie in the Chicago metropolitan area— did not early-certify. Illinois’s early-certified data showed a violating monitor in the Chicago area. At the first step of the Chicago-area designation process, the EPA relied on Illinois’s early-certified data, noted the violation, and thus proceeded to the second step’s multi-factor contribution analysis for all Chicago-area counties.

Mississippi argues that, because the EPA only possessed early-certified data from Illinois, it used a mismatched data-set for Chicago’s designations. Consequently, Mississippi claims that the EPA took different approaches to dataset selection between Memphis and Chicago. Mississippi’s argument rests on a flawed understanding of the EPA’s designation process.

At the first step of the process, a single violating monitor suffices to conclude the analysis and move to the second step. Though only Illinois had early-certified its *161data, that data showed a violating monitor. That was enough to terminate the first step of the process and move to the second step. It thus became irrelevant whether Wisconsin or Indiana data showed any violations: the EPA would proceed to the second step of the analysis regardless, based on the Illinois violation alone. The EPA therefore had a sufficient matched dataset of 2009 to 2011 data (albeit data from only one state, Illinois) to proceed to the second step of the designation process using 2009 to 2011 data alone. By contrast, the EPA had no matched dataset of 2009 to 2011 data in the Memphis area sufficient to complete the first step of the two-step process using that data alone. While data showing a single violating monitor are enough to end the first step and proceed to the second step, data showing all monitors in compliance would be needed to avoid proceeding to the second step’s multi-factor analysis — ie., to terminate the two-step process at the first step.

As a result, when Arkansas declined to early-certify its 2011 data, the EPA could not determine if the. entire Memphis CBSA showed NAAQS compliance at all monitors for the 2009 to 2011 period; the agency lacked a sufficient 2009 to 2011 matched dataset with which to do so. The EPA then relied on the most recent matched dataset sufficient to complete the first-step analysis (the 2008 to 2010 data), just as the EPA selected the most recent matched dataset sufficient for the first-step analysis of the Chicago area. The EPA therefore acted in a consistent manner in both areas, each time using the most recent matched datasets sufficient to complete the first step of the two-step designation process.

3. Challenge to the Second Step of the Designation Process

Mississippi also challenges the EPA’s application of the second step of the designation process. The EPA acted arbitrarily, the state argues, in applying the multi-factor test and concluding that De-Soto County contributed to the Shelby County violation. We find no reason to disturb the EPA’s analysis.

First, Mississippi challenges the EPA’s differing articulations of the multi-factor test. As pronounced in the 2008 Guidance, the EPA originally conceived of that test as consisting of nine factors. In making the final designations, the EPA applied a five-factor test. See supra § I.B-C, The state argues that the EPA’s “consolidation]” of the test from nine to five factors was arbitrary and capricious. State & County Br. 15. We disagree.'

At the outset, we do not necessarily agree that the EPA was required to adhere to the 2008 Guidance. The 2008 Guidance did not purport to be a legislative rule, and it explicitly provided that it was “not binding on states, tribes,' the public or the EPA.” 2008 Guidance at 4; cf. Catawba Cnty., 571 F.3d at 33-34 (materially similar guidance for PM2,5 NAAQS designations did not “create or modify legally binding rights”).

But even if we assume that the 2008 Guidance was binding, the EPA did not deviate from it in the final designations. The “consolidation” of the factors was just that — a consolidation.. It effected no deletion. During the final designation process, the agency simply grouped several of the 2008 Guidance factors into a single factor, the consideration of which necessarily entailed consideration of the multiple 2008 Guidance factors now residing within it. We find no examples of a final designation that failed to consider a factor identified in the 2008 Guidance. With “no bright line for any of the factors,” and with each factor “weighted considering the unique circumstances of each nonattainment area,” Response to Comments at 61, the consolidation worked no substantive *162change and thus affords no basis for setting aside the EPA’s analysis.

Second, Mississippi challenges the EPA’s specific application of the multi-factor test to DeSoto County. We accord the EPA “extreme deference” in applying that test, and will overturn the EPA’s designations only if the agency applied the test “so erroneously in a particular case that it could not have reasonably concluded that a county was contributing to nearby violations.” Catawba Cnty., 571 F.3d at 40-41. This is not such a case. The agency provided data showing that DeSoto County’s NOx and S02 (ozone precursors) emissions were the second-highest in the Memphis CBSA. Memphis Area Designations at 8. The county also had the second highest number of workers commuting to counties with violating monitors, the second highest number of vehicle miles traveled in the CBSA, and the highest percentage population growth over the last decade. Those factors led the EPA to conclude that DeSoto County was integrated with Shelby County in a way that indicated ozone contribution. Id. at 9-10. Additionally, meteorological analysis at the Shelby County monitor showed weather patterns characterized in part by winds blowing in from DeSoto County. Id. at 12. On those bases, the EPA reasonably concluded that DeSoto County contributed to the Shelby County violation.

Mississippi principally argues that significant “commerce activity” occurring outside of DeSoto County (including interstate highway traffic, rail and barge transportation, diesel fuel sales, and air traffic) means that other' counties contribute to the Shelby County violation more than DeSoto County does — and that, because some of those counties avoided nonattainment designations, DeSoto County should, too. Miss. Dep’t of Envtl. Quality, Air' Div., 2008 Ozone Standard Designation Recommendation for DeSoto County, Mississippi 8-12 (Feb. 2012). But the EPA considered that argument and determined in a well-reasoned analysis that the data from Mississippi was only one consideration in the designation process. See Response to Comments at 97; see also Memphis Area Designations 1-31. The EPA concluded that DeSoto County did contribute to Shelby County’s violation in light of the many other factors the agency considered. Memphis Area Designations at 16. .

Looking at the same data, Mississippi would simply reach a different conclusion. We, however, do not sit to second-guess the EPA’s conclusions in an area identified by the Congress as. within the agency’s technical expertise. We only ask if the EPA “considered all relevant factors and articulated a rational connection between the facts found and the choice made.” ATK Launch Sys., 669 F.3d at 336 (internal quotation marks omitted). We conclude that it did.

With that conclusion, and having considered Mississippi’s other challenges and determined that they lack merit, we deny the state’s petition for review. See Catawba Cnty., 571 F.3d at 52.

E. Lake & Porter Counties, Indiana

Petitioner Indiana challenges the designation of two of its counties as nonat-tainment. According to Illinois’s certified 2009 to 2011 data, the monitoring site at Zion, Illinois exceeded the NAAQS by 1 part per billion (ppb). See Chieago-Na-perville, Illinois-Indiana-Wisconsin Area Designations for the 2008 Ozone NAAQS at 7-8 [hereinafter Chicago Area Designations]. Zion is about sixty miles from the Indiana border and, like the Indiana counties at issue here, belongs to the Chieago-Naperville-Michigan City CSA. Following the 2008 Guidance, the EPA presumed that all counties in this CSA should be *163designated as nonattainment areas due to the Zion violation, and then conducted its five-factor analysis. The agency preliminarily concluded that three Indiana counties — Lake, Porter, and Jasper — should be included in the nonattainment area.

In response to the EPA’s 120-day letter, Indiana pointed to multiple asserted flaws in the EPA’s analysis. Most relevant here, it said that the agency had failed to account for the impact of a recent statutory change to Illinois’s vehicle emissions testing program. It also maintained that the agency’s meteorological analysis suffered from multiple weaknesses and inconsistencies.

The EPA ultimately reversed its designation of Jasper County, but finalized the nonattainment designations of Lake and Porter Counties. Chicago Area Designations at 21. Indiana now challenges those nonattainment designations as arbitrary and capricious.

1. Challenge Regarding Illinois’s Vehicle Inspection Change

First, Indiana challenges the EPA’s position regarding Illinois’s statutory change. After a prior nonattainment designation, Illinois had established a vehicle inspection and maintenance program that covered all model years beginning in 1968.11 In 2006, however, Illinois exempted vehicles with model years between 1968 and 1995 from the testing requirements. See 625 Ill. Comp. Stat. 5/13C — 15(a)(6)(L) (2012). Indiana maintains that it was the increase in vehicle emissions accompanying this exemption that directly caused the violation at the Zion monitor. Moreover, it contends that this legislative change amounted to an intentional violation of Illinois’s SIP.

As the EPA points out, we made clear in Catawba County that a “contributing” county need not be the but-for cause of a violation in order to warrant a nonattainment designation. Resp’t’s Br. 94; see Catawba Cnty., 571 F.3d at 39 (“[E]ven were we to think that ‘contribute’ unambiguously means ‘significantly contribute,’ we still disagree that ‘significantly contribute’ unambiguously means ‘strictly cause.’ ”). And here, regardless of Illinois’s statutory change, the EPA’s five-factor analysis demonstrated that both Lake and Porter Counties contributed to the Zion monitor. Chicago Area Designations at 6-21.12

The alleged illegality of Illinois’s statutory change does not affect our conclusion. The Clean Air Act offers other avenues for addressing a State’s failure to comply with its SIP. In particular, the EPA Administrator can call for a SIP revision after “findfing] that the applicable implementation plan for any area is substantially inad*164equate” to comply with the NAAQS. 42 U.S.C. § 7410(k)(5). The EPA declined to do so here and, instead, recently approved the Illinois change.13 Indiana has since petitioned the Seventh Circuit to review the EPA’s approval. See EPA 28(j) Letter (Oct. 22, 2014). That is the appropriate forum for challenging the Illinois change, which in no way diminished the contribution of the Indiana counties.

2. Challenge to the EPA’s Response to Comments

Next, Indiana argues that the EPA failed to adequately respond to its comments about the impact of Milwaukee, Wisconsin’s emissions on the violation at the Zion monitor. According to the source apportionment modeling submitted by Indiana, the Milwaukee area contributed over 5 ppb to the Zion violation, while Lake, Porter, and Jasper Counties contributed 4 ppb, 2 ppb, and 0.5 ppb, respectively. See Letter from Ind. Dep’t of Envtl. Mgmt. to EPA, Enclosure 1 at 13-14 (Apr. 13, 2012). This, Indiana maintains, produced • the “inconsistent and unfounded” result of nonattainment designations for the Indiana counties but an attainment designation for the Milwaukee area. Id. at 14.

As an initial matter, we note that, because the Milwaukee area is not a single county but rather is a metropolitan area made up of five counties, Indiana’s argument is premised on an apples-to-oranges comparison. More important, we have no basis for finding the EPA’s designations inconsistent given that Indiana’s modeling — which was limited to meteorological linkages and therefore fell short of a full analysis — did not establish that Milwaukee “contributed to” the Zion- violation under the agency’s five-factor analysis. By contrast, after conducting its full five-factor analysis, the EPA found that Lake and Porter Counties did contribute. Accordingly, the EPA’s determination regarding the Milwaukee metropolitan area was neither unreasonable nor inconsistent with its determination regarding the Indiana counties.

We also find that the EPA did adequately respond to Indiana’s comments about its modeling results, although without mentioning Milwaukee specifically. Indeed, the modeling was one of the factors that led the EPA to reconsider its designation of Jasper County. See Chicago Area Designations at 21 (describing Jasper County’s 0.5 ppb contribution as “not significant”). But the EPA simply disagreed with Indiana’s premise that 2 ppb and 4 ppb were insufficient contributions when considered as part of the five-factor test, for reasons that were reasonable and well explained. See id. at 18 (“In keeping with EPA’s ozone contribution levels used to select states that should be covered in regional emission control programs, 2 ppb to 4 ppb ozone concentration contributions are considered to be significant ozone contributions.”).

3. The Remaining Challenges

Finally, we briefly consider Indiana’s remaining arguments. First, the record does not support Indiana’s claim that the EPA improperly relied on late-submitted data from Wisconsin’s Chiwaukee Prairie monitor, rather than relying solely on the Zion monitor data, in making the contribution determinations regarding the Indiana counties. See Chicago Area Designations at 8 (noting that the EPA considered the Wisconsin data in determining whether Kenosha County, Wisconsin (and not the *165Indiana counties) should be included in the Chicago nonattainment area); id. at 21-22 (describing bases for Lake, Porter, and Kenosha County designations). Second, the EPA did not fail to adequately explain why it used some 2006 to 2008 weather data in conducting the contribution analysis. The agency explained that historical data provided a “general conceptual model to explain the development and transport of high ozone levels in this area.” Addendum to Response to Comments at 7 (May 31, 2012); see also EPA Response to Indiana Pet. for Reconsideration 3. That explanation is deserving of the deference that we give to the EPA’s “evaluati[on] [of| scientific data within its technical expertise,” Catawba Cnty., 571 F.3d at 41 (quoting City of Waukesha, 320 F.3d at 247).

In sum, we reject Indiana’s contention that the EPA’s designations of Lake and Porter Counties are arbitrary or capricious.

F. Wise County, Texas

Petitioners State of Texas; Wise County, Texas; Texas Commission on Environmental Quality; Devon Energy Corporation; Targa Resources Corporation; the Texas Pipeline Association; and the Gas Processors Association (collectively, Texas Petitioners) challenge the EPA’s designation of Wise County as nonattainment. They make several claims, including that the EPA subjected Wise County to arbitrarily disparate treatment, violated the U.S. Constitution and acted beyond its authority under the Clean Air Act. For the reasons discussed below, however, we do not disturb Wise County’s nonattainment designation.

1. Wise County Background

Wise County is one of 22 counties in and around the Dallas-Fort Worth metropolitan area, which reports some of the most severe NAAQS violations in the country. Although Wise County has no monitor of its own, it borders several counties with a total of seven violating monitors, the closest of which reports ambient ozone levels that exceed the 2008 NAAQS by 0.010 ppm. Moreover, because Wise County falls vpthin the CSA of Dallas-Fort Worth, it is presumptively included within the nonattainment area.

Despite Wise County’s presumptive inclusion in the Dallas-Fort Worth nonat-tainment area, the EPA designated it as attainment when it updated the ozone NAAQS in 1997. For this reason, Texas did not include Wise County among the nine Dallas-Fort Worth counties it recommended for nonattainment status when it submitted its initial designations to the EPA in March 2009.14 On December 9, 2011, the EPA informed Texas that it planned to include Wise County in the Dallas-Fort Worth nonattainment area due to its “comparatively high emissions” and “close proximity ... to violating monitors.” See Texas Area Designations for the 2008 Ozone NAAQS at 13 [hereinafter Preliminary Dallas-Fort Worth Area Designations].

The EPA redesignated Wise County based on the five-part “weight of the evidence analysis” articulated in the 2008 Guidance.15 See id. at 1-2. The second and third factors — emissions data and meteorology — factored prominently in the EPA’s decision. See id. at 13. As for emissions, the EPA concluded that oil-and-*166gas collection and production in the Barnett Shale reservoir — a gas-rich geological formation covering a significant portion of Wise County^ — -resulted in Wise County’s inclusion among the eight highest emissions-producing counties in the Dallas-Fort Worth area.16

As for meteorology, although historic wind patterns in the Dallas-Fort Worth area suggest that air does not normally move from Wise County to counties with monitors registering NAAQS violations, the EPA concluded that Wise County was upwind of the monitors on days when ozone levels at the monitors peaked. See Preliminary Dallas-Fort Worth Area Designations at 10. In reaching this conclusion, the EPA used the National Oceanic and Atmospheric Administration’s Hybrid Single Particle Lagranian Integrated Trajectory (HYSPLIT) model instead of relying solely on historic wind patterns in the Dallas-Fort Worth area. See id. HYS-PLIT charts the path, or “back trajectory,” that air takes before it collects in a certain area. See id. According to the EPA, HYSPLIT modeling “is specifically designed to give an estimate of the probable path a parcel of air travels in reaching a given location at a given time” and is particularly illuminating for an area like Wise County, which has “light and variable” wind patterns. Response to Comments at 59-60.

After the EPA notified Texas that it planned to include Wise County in the Dallas-Fort Worth nonattainment area, numerous individuals and organizations submitted comments urging the EPA to reconsider its Wise County designation. One commenter insisted that other Texas counties were more responsible than Wise County for the NAAQS violations in the Dallas-Fort Worth area. Others argued that the EPA’s use of HYSPLIT modeling was arbitrary and capricious because, when designating other areas of the country, the EPA relied solely on historic wind patterns. According to these commenters, if the EPA had done the same with Wise County, it would not have designated Wise County as nonattainment because, according to historical wind patterns in the Dallas-Fort Worth area, Wise County was downwind of violating monitors more than 95 per cent of the time.

For its part, Petitioner Texas Commission on Environmental Quality (Texas Commission) submitted its own data based on photochemical grid source apportionment modeling. Source-apportionment modeling helps determine the potential future impact of an emissions source area (such as Wise County) on downwind monitors by “keepfing] track of the origin of the [ozone] precursors creating the ozone.” Industrial Br. 7. It does so by combining “the meteorology/transport of air parcels during high ozone days with the emissions of [a] specific area[ ],” (here, Wise County), “to evaluate potential impact on ozone levels.” Dallas-Fort Worth, Texas Final Area Designations for the 2008 Ozone NAAQS at 16 [hereinafter Final Dallas-Fort Worth Area Designations]. Although the EPA does not typically perform source-apportionment modeling during the NAAQS designation process, it “has used it in the past for large-scale rulemakings, such as the Clean Air Interstate Rule and Cross State Air Pollution Rule” and it considers source-apportionment modeling data if a state submits it. See Resp’t’s Br. 126. According to the Texas Petitioners, source-apportionment modeling suggests that Wise County emissions had only a negligible impact on the monitors register*167ing NAAQS violations in the Dallas-Fort Worth area.

On April 30, 2012, the EPA issued its omnibus Response to Comments, many of which addressed the objections to the Wise County designation. The EPA defended HYSPLIT modeling as an “excellent toolE ]” that it generally “preferfs] over more basic assessments of wind speed and direction.” Response to Comments at 59. The EPA found HYSPLIT modeling to be a more precise measure of wind patterns than historic data, which data, according to the agency, is “potentially misleading in cases where wind speeds are light and variable, or vary substantially across the location of the meteorological observation and the monitored high ozone concentrations.” Id. These conditions existed in the Dallas-Fort Worth area.17 Although the EPA acknowledged it could not always use HYSPLIT modeling, it nonetheless declined to ignore HYSPLIT data “where the information is available, even if the information is not available in all areas.” Response to Comments at 59.

Along with its omnibus responses, the EPA issued its Final Dallas-Fort Worth Area Designations, which again applied the five-factor test. In that document, the EPA addressed the source-apportionment modeling submitted by the Texas Commission. The EPA took issue with the model’s methodology and made several amendments to it.

First, the EPA faulted the Texas Commission for not using data from an entire ozone season in its model. To account for this omission, the EPA examined not only the average (ie., relative) impact of Wise County emissions on Dallas-Fort Worth monitors but also the absolute (ie., maximum) impact of the emissions. See Final Dallas-Fort Worth Area Designations at 17. The average/relative approach advocated by the Texas Commission averaged the impact that Wise County emissions might have on the monitors on all days when the monitors were expected to exceed the ozone NAAQS. As a practical matter, averaging the impact of Wise County emissions meant that the Texas Commission’s model accounted for days on which wind patterns were not expected to move air pollutants from Wise County to the violating monitors. According to the EPA, the Texas Commission’s average approach had “the effect of masking the impacts that occur on days when the wind does flow from Wise County to violating monitors,” an imprecision that was aggravated by the model’s limited dataset. See Resp’t’s Br. 136 (emphasis added). To account for this imprecision, EPA chose to look at the “direct,” or “absolute,” predicted effect that Wise County emissions would have on violating monitors rather than the average effect they were expected to have.

Second, the EPA noted that the Texas Commission’s source-apportionment model under-predicted peak ozone levels in the Dallas-Fort Worth- area by a range of 0.005 to 0.020 ppm. As a practical matter, the under-prediction meant that the Texas Commission’s model underestimated the number of days that Wise County contributed to NAAQS violations. To compensate therefor, the EPA examined the impact of Wise County emissions not only on days when the monitors exceeded the ozone NAAQS threshold of 0.075 ppm, but also on days when the monitors reported ozone levels in excess of 0.070 ppm.

*168After making these adjustments, the EPA reinterpreted the data from the Texas Commission’s source-apportionment model and concluded that it in fact supported including Wise County in the Dallas-Fort Worth nonattainment area. See Final Dallas-Fort Worth Area Designations at 20. Specifically, the EPA concluded that Wise County emissions (1) “resulted in 6 occurrences (over 4 days) of an impact of more than 0.75 ppb days” on Dallas-Fort Worth area monitors; (2) “had even larger impacts of up to 5 ppb on the Eagle Mountain Lake monitor,” a monitor one-half mile from the Wise County border that reported particularly severe NAAQS violations; and (3) “resulted in 9 occurrences (over 5 days) [causing] impacts of more than 0.75 ppb [to] oceur[] at” Dallas-Fort Worth monitors. See id. For these reasons, the EPA maintained its inclusion of Wise County in the Dallas-Fort Worth nonattainment area.

Dozens of individuals and organizations filed petitions for reconsideration of the EPA’s Wise County nonattainment designation, including the Texas Commission and the other Texas Petitioners. On December 14, 2012, the EPA denied each petition for reconsideration. Before us, the Texas Petitioners’ challenges to the EPA’s Wise County designation are grouped as follows: (1) The EPA’s use-of HYSPLIT Modeling and its re-evaluation of the Texas Commission’s source-apportionment modeling were arbitrary and capricious; (2) the EPA’s designation of Wise County as nonattainment violated the Commerce Clause, U.S. Const, art. I, § 8, cl. 8, the Tenth Amendment, id. amend. X, and the Due Process Clause, id. amend. V; and (3) the EPA violated at least one of several statutory provisions, including provisions of the Clean Air Act. We address each argument in turn.

2. The Arbitrary & Capricious Challenges

The Texas Petitioners’ primary arguments are that the EPA erred when it (i) used HYSPLIT modeling rather than prevailing wind patterns18 and (ii) adjusted the Texas Commission’s source-apportionment modeling.19 To prevail on either argument, the Texas Petitioners must demonstrate that the EPA acted arbitrarily and capriciously and, to do that, they must show that the EPA either failed to consider “all relevant factors” or to articulate a “rational connection between the facts found and the choice made.” ATK Launch Sys., 669 F.3d at 336. Mindful of the “extreme degree of deference” we owe to the EPA “when it is evaluating scientific data within its technical expertise,” Catawba Cnty., 571 F.3d at 41, and for the reasons stated below, we conclude that neither argument has merit.

i. HYSPLIT Modeling

The Texas Petitioners challenge the EPA’s use of HYSPLIT modeling on three fronts. First, they argue that the EPA could not legitimately use HYSPLIT modeling at all because HYSPLIT “cannot measure ozone formation or transport.” State & County Br. 45. Second, they contend that the EPA arbitrarily treated Wise County differently by using HYSPLIT modeling to designate it as nonattainment while using historic wind patterns to designate other allegedly similar counties as attainment. And third, they argue that, even among other counties that the EPA subjected to HYSPLIT modeling, it arbitrarily treated Wise County worse because the respective HYSPLIT models demonstrated that wind moved through those other counties — each of which the EPA designated as attainment — more frequent*169ly than it moved through Wise County. We address each argument in turn.

First, we find no merit in the Texas Petitioners’ conclusory argument that the EPA erred by using HYSPLIT modeling at all because HYSPLIT modeling “cannot measure ozone formation or transport.” See State & County Br. 45-46. Indeed, we rejected a materially indistinguishable challenge in ATK Launch Systems, 669 F.3d at 339, a case involving the EPA’s 2006 fine particulate matter NAAQS designations. See id. at 334. We did so there because the EPA had taken “reasonable steps to ensure that the ‘HYSPLIT’ model’s limitations were considered.” Id. at 339 (quotation mark omitted).

Here too, the EPA took reasonable steps to account for HYSPLIT’s limitations by evaluating the source-apportionment modeling and historical wind data that the Texas Commission submitted during the comment period. See Final Dallas-Fort Worth Area Designations at 14-20, 23. Because “[o]zone and ozone precursors can be transported to an area from sources in nearby areas or from sources located hundreds of miles away,” see 2008 Designations Rule, 77 Fed.Reg. at 30,088, the EPA reasonably concluded that HYS-PLIT modeling, as a more precise measurement of the path taken by air masses containing ozone precursors, was useful in determining whether wind moving through Wise County could have transported emissions to the areas with the violating monitors.

Second, we find no merit in the Texas Petitioners’ argument that the'EPA’s use of HYSPLIT modeling to designate Wise County as nonattainment amounts to arbitrarily disparate treatment. At the outset, it bears repeating that this Court has expressly sanctioned the EPA’s use of a holistic, multi-factor, totality-of-the-circumstances test for making NAAQS determinations, see ATK Launch Sys., 669 F.3d at 336; Catawba Cnty., 571 F.3d at 39, and we have twice iterated that, when using a multi-factor test, “‘discrete data points’ are not determinative” because isolating any one discrete consideration “ ‘ignores the very nature of the ... test, which is designed to analyze a wide variety of data on a case-by-case basis.’.” ATK Launch Sys., 669 F.3d at 336 (quoting Catawba Cnty., 571 F.3d at 39) (emphasis added; alteration omitted). Indeed, because the EPA’s “holistic assessment of numerous factors ... drives the process,” we have recognized that “no single factor determines a particular designation.” Id. For this reason, the EPA could have subjected Wise County to arbitrarily disparate treatment only if it treated genuinely “similar counties ” dissimilarly. Id. (emphasis in original). Given “significant” differences among counties, “a direct one-to-one comparison of the data,” including the methods used to measure such data, could be “inappropriate” or even “illogical.” Id. at 337.

As noted, the EPA conducted a HYS-PLIT analysis in areas where it “believed [HYSPLIT] could provide additional insight into whether [the] area[] contribute[s] to nonattainment.” Resp’t’s Br. 110 n. 47. The EPA reasonably determined that Wise County was one such area because Dallas-Fort Worth “experiences light wind speeds and winds from variable directions,” making HYSPLIT’s more sophisticated evaluation of wind patterns “a more useful tool than annualized wind patterns.” EPA Response to Pet. for Reconsideration from Devon Energy Corp. at 12. According to the EPA, this more refined analysis was not necessary for all areas of the country, particularly those in which “there was not significant debate over whether [they] should be included” in a nonattainment area. See Resp’t’s Br. 111. The EPA’s decision to use HYSPLIT analysis in one area but not in another fits *170comfortably within the agency’s “technical expertise,” Catawba Cnty., 571 F.3d at 41, and the EPA’s explanation for the differing treatment was rational.

Moreover, although the Texas Petitioners direct this Court to other attainment areas that were not evaluated using HYS-PLIT modeling — specifically, Orange County and Cattaraugus County in New York — the “significant” differences between Wise County and those counties “make a direct one-to-one comparison of the data underlying the analyses inappropriate.” ATK Launch Sys., 669 F.3d at 337. For instance, the EPA justified its Orange County attainment designation, in part, on its finding that “the density of [Orange County’s] emissions and vehicle usages are not of the level of the other counties in the CSA that are in New York’s proposed New York-Northern New Jersey-Long Island, NY-NJ-CT nonat-tainment area.” New York-Northern New Jersey-Long Island, NY-NJ-CT Nonattainment Area Designations for the 2008 Ozone NAAQS at 16 (emphasis added). In contrast, the EPA justified its nonattainment designation of Wise County, in part, based on the “[t]he close proximity of [Wise County’s] comparatively high emissions to violating monitors.” Final Dallas-Fort Worth Area Designations at 23 (emphasis added).

Similarly, the EPA designated Cattarau-gus County as attainment not only because “it is in the prevailing downwind direction from” the nearest violating monitor but also because “other monitors representative of Cattaraugus County, as well as the rest of upstate New York, are attaining the ozone standard.” See Attainment Status for Jamestown, New York and the Remainder of Upstate New York at 6 (emphasis added). But in the Dallas-Fort Worth area, seven violating monitors surrounded Wise County and some of the monitors — including one located one-half mile from Wise County’s border — reported levels of ambient ozone higher than anywhere else in the United States. Because “the core reason for the disparate designations” did not, as the Texas Petitioners would have it, reflect an “inconsistent approach to meteorology,” Industrial Br. 19, the EPA did not arbitrarily and capriciously treat Wise County differently by evaluating its wind patterns using HYSPLIT modeling instead of prevailing wind patterns.

Third, when Wise County is compared to other counties for which the EPA used HYSPLIT modeling, it is clear that the EPA did not arbitrarily subject Wise County to disparate treatment. The Tex-as Petitioners point to four other counties — York, Dauphin and Lawrence Counties in Pennsylvania and Roane County, Tennessee — each of which the EPA designated as attainment notwithstanding HYS-PLIT modeling demonstrated that air moved through them to violating monitors more frequently than through Wise County. But again, a holistic look at why the EPA designated these counties attainment but designated Wise County nonattainment demonstrates that the EPA did not act arbitrarily or capriciously.

For example, York and Dauphin Counties are both near Lancaster County, which houses all violating monitors in the area. Because Lancaster County “is served by a single-county transportation-planning agency,” the EPA concluded that there were “strong jurisdictional arguments” for designating Lancaster as “a single county nonattainment area” and, accordingly, designating all other counties in the vicinity — including York and Dauphin — as attainment. See Pennsylvania Area Designations for the 2008 Ozone NAAQS at 29-31. In contrast, Wise County is part of the Dallas-Fort Worth CSA (which means it is presumptively in-*171eluded in the Dallas-Fort Worth nonat-tainment area) and is also part of the Dallas-Fort Worth metropolitan planning organization (which implements programs and projects to reduce emissions across all included counties). In other words, jurisdictional and regional planning concerns— not differing approaches to HYSPLIT modeling data — drove the EPA’s conclusion that York and Dauphin Counties should be designated as attainment while Wise County should be designated as non-attainment.

The Texas Petitioners’ comparisons of Wise County to Roane County, Tennessee, and Lawrence County, Pennsylvania, fare no better. Roane County is “geographically separated from the nearest county with a violating monitor” by approximately thirty miles and the ozone levels in the county between Roane and the next county with a violating monitor are in attainment. Resp’t’s Br. 122. The monitor in Lawrence County reports ozone levels that, at 0.066 ppm, are well below the EPA’s NAAQS 0.075 ppm threshold. Moreover, the county with a violating monitor nearest to Lawrence County — Allegheny County— is not adjacent to Lawrence County. In contrast to both Roane County and Lawrence County, Wise County is adjacent to multiple counties reporting severe NAAQS violations, the closest of which is located a mere half mile from the Wise County line.

The dispositive principle that the Texas Petitioners try to, but ultimately cannot, avoid is that under the EPA’s holistic analysis, “discrete data points” like the data from HYSPLIT modeling “are not determinative, because elevating them ignore[s] the very nature of the [holistic] test, which is designed to analyze a wide variety of data on a case-by-case basis.” ATK Launch Sys., 669 F.3d at 336 (quotation mark omitted). Based on the foregoing analysis, we cannot say that, had the EPA declined to evaluate Wise County’s wind patterns using HYSPLIT modeling, Wise County “would not have been designated nonattainment.” Industrial Br. 19. Because none of the areas discussed by the Texas Petitioners is truly “similarly situated” to Wise County, and because the EPA fully and rationally supported its use of HYSPLIT modeling for Wise County, it did not act arbitrarily or capriciously.

ii. Source-Apportionment Modeling

The Texas Petitioners also challenge the EPA’s modification of the Texas Commission’s source-apportionment modeling on three fronts. First, they argue that the EPA has not rationally explained why it considered the source-apportionment modeling’s projected absolute impact — instead of its projected relative impact — that wind from Wise County would have on violating Dallas-Fort Worth area monitors. Second, they argue that the EPA’s analysis of the Texas Commission’s source-apportionment modeling was inconsistent with its analysis of source-apportionment modeling submitted in connection with Illinois’s designation of Lake County. And third, they argue that the EPA’s decision to examine the model’s projected absolute impact rather than its relative impact violated the EPA’s earlier modeling guidance.

We note, at the outset, that the EPA’s application, interpretation and modification of source-apportionment modeling plainly fall “within its technical expertise” and thus we owe it “an extreme degree of deference.” ATK Launch Sys., 669 F.3d at 338 (quotation marks omitted). To withstand judicial review, the EPA needs to articulate only a “rational connection between the facts found and the choice made,” Burlington Truck Lines, 371 U.S. at 168, 83 S.Ct. 239, show that it treated “similar counties ” similarly, ATK Launch Sys., 669 F.3d at 336 (emphasis in origi*172nal), and demonstrate that it did not run afoul of binding guidance, see generally Appalachian Power Co. v. EPA, 208 F.3d 1015, 1020-23 (D.C.Cir.2000). Because the EPA has done all three, we will not disturb its designation of Wise County as nonat-tainment based on the Texas Petitioners’ objections to its interpretation of the Tex-as Commission’s source-apportionment modeling.

First, the Texas Petitioners challenge the EPA’s decision to reinterpret the source-apportionment modeling submitted by the Texas Commission. As discussed, supra § III.F.l, when the EPA received the Texas Commission’s source-apportionment modeling data during the comment period, it observed that the model did not rely on data from an entire ozone season. Rather, the projections in the Texas Commission’s model relied on data from June 2006 only. The Texas Commission based its approach on the fact that June 2006 purportedly presented “an exceptionally rich set of air quality and meteorological measurements,” “had the most high-ozone days of any month” and experienced “all the meteorological conditions linked to formation of high ozone concentration.” See Response to Texas Commission on Environmental Quality’s Reconsideration Pet. at 3.

Despite these assurances, the EPA did not agree that one month of data, even an “exceptionally rich” month, was sufficient. Specifically, the EPA observed that the ozone season in the Dallas-Fort Worth area was bimodal (ie., reporting its highest ozone values in July-September but experiencing a lower ozone peak in May-June) and that the Texas Commission’s reliance on limited data meant that it failed to account for “all of the meteorology regimes conducive for ozone events” in the Dallas-Fort Worth area. See Final Dallas-Fort Worth Area Designations at 16. According to the EPA, “emphasis on the average modeled impact is more appropriate when a full ozone season of model results is available.” See Resp’t’s Br. 131. Because the Texas Commission’s model was premised on baseline data excluding “events that happen in mid to late-summer that often set” the Dallas-Fort Worth area’s ozone levels, the EPA examined both the projected average impact and the projected maximum impact of Wise County emissions. See Final Dallas-Fort Worth Area Designations at 16.

At bottom, the EPA had a “basic obligation” to conduct “reasoned decisionmak-ing.” Catawba Cnty., 571 F.3d at 25. When presented with the Texas Commission’s source-apportionment modeling, the EPA determined that it “needed to be carefully evaluated and could not simply be accepted at face value,” Resp’t’s Br. 126, identified several methodological flaws in the Texas Commission’s data, adjusted the Texas Commission’s submissions to account for the flaws and articulated, quite thoroughly, a “rational connection between the facts found and the choice made.” Burlington Truck Lines, Inc., 371 U.S. at 168, 83 S.Ct. 239. On this record, we cannot say that the EPA acted arbitrarily or capriciously in re-evaluating the Texas Commission’s source-apportionment modeling data. Rather, the EPA’s thorough treatment of all available data indicates that it in fact “surpassed” its “obligation of reasoned decisionmaking.” Catawba Cnty., 571 F.3d at 25.

Second, the Texas Petitioners argue that the EPA’s modification to the Texas Commission’s source-apportionment modeling subjected Wise County to arbitrarily disparate treatment. They compare the EPA’s interpretation of the Texas Commission’s modeling to its interpretation of source-apportionment modeling for the Chicago area. Specifically, they argue that (1) emissions from Jasper County, a Chica*173go-area county with attainment status, had a projected average impact on violating monitors similar to Wise County’s; (2) the EPA should have evaluated the average impact of Wise County’s emissions on violating monitors as it did for Jasper County; and (3) the EPA’s evaluation of Wise County’s maximum, as opposed to relative, estimated impact was, accordingly, inconsistent and resulted in an arbitrarily different result between Wise County and Jasper County.

Again, we emphasize that applying different methods to different areas, standing alone, does not give rise to arbitrarily disparate treatment and given “significant” relevant differences between two areas, “a direct one-to-one comparison of the data” or the methods used to measure such data can be “inappropriate.” ATK Launch Sys., 669 F.3d at 337. Here, the significant difference lies in the quality of data submitted by the Texas Commission compared to that submitted in support of Jasper County. Specifically, the source-apportionment model submitted in support of the Chicago-area designations included data from a full ozone season, which made “emphasis on the average modeled impact ... more appropriate.” Resp’t’s Br. 131. As noted, the EPA modified the Texas Commission’s source-apportionment model because it did not include data from a full ozone season.

Moreover, the EPA had to compensate for the fact that the Texas Commission’s source-apportionment model underestimated the number of days that monitors in the Dallas-Fort Worth area exceeded the ozone NAAQS because the model under-predicted peak ozone levels around the monitors, sometimes by a significant range. The source-apportionment model for Jasper County, however, had the opposite problem; it did not account for recent emissions reductions at a Jasper County power plant and thus the Chicago-area source-apportionment model over-reported Jasper County’s emissions impact. See Chicago Area Designations at 9-10. Stated differently, because Wise County’s model under-reported its emissions impact and Jasper County’s model over-reported its emissions impact, the EPA reasonably concluded that the two counties should receive different attainment designations.

Third, the Texas Petitioners argue that the EPA arbitrarily and capriciously deviated from its earlier guidance on source-apportionment modeling, which guidance allegedly expressed a preference for relative, rather than absolute, modeling. Specifically, they argue that the EPA’s reb-anee on Wise County’s maximum potential emissions impact directly conflicts with the EPA’s 2007 “Guidance on the Use of Models and Other Analyses for Demonstrating Attainment of Air Quality Goals for Ozone, PM2.6, and Regional Haze” (2007 Attainment Guidance). In that guidance, the EPA stated that its “recommended test is one in which model estimates are used in a ‘relative’ rather than ‘absolute’ sense.” Id. at 15.

As a threshold matter, the 2007 Attainment Guidance does not speak to the use of source-apportionment modeling in the designation process; rather, it recommends procedures that a state can use after it has been designated as nonattainment to show that its proposed emission control strategy will eventually result in attainment status. But even assuming that the 2007 Attainment Guidance informs the current NAAQS designation process, the EPA did not err by deviating from it. Indeed, the 2007 Guidance expressly contemplates deviations in appropriate cases:

This document does not substitute for any Clean Air Act provision or EPA regulation, nor is it a regulation itself. Thus, it does not impose binding, en*174forceable requirements on any party, nor does it assure that EPA will approve all instances of its application. The guidance may not apply to a particular situation, depending upon the circumstances. The EPA and State decision makers retain the discretion to adopt approaches on a case-by-case basis that differ from this guidance where appropriate ....

Users are cautioned not to regard statements recommending the use of certain procedures or defaults as either precluding other procedures or information, or providing guarantees that using these procedures or defaults will result in actions that are fully approvable.... EPA cannot assure that actions based upon this guidance will be fully approvable in all instances.

2007 Attainment Guidance at ix.

As noted, the EPA fully explained why it revised and independently evaluated the Texas Commission’s source-apportionment modeling to account for “the limited data set [the Texas Commission] relied upon.” Resp’t’s Br. 136. Because the 2007 Attainment Guidance did not compel the EPA to limit its consideration to relative projected impacts, and because the EPA articulated a “rational connection between the facts found and the choice made,” Catawba Cnty., 571 F.3d at 41, it did not act arbitrarily or capriciously when it relied on Wise County’s absolute, rather than relative, impact on NAAQS-violating monitors.

The fundamental deficiency in the Texas Petitioners’ challenges to the EPA’s revision of the Dallas-Fort Worth area source-apportionment model is that, to establish that “EPA’s administration of the complicated provisions of the Clean Air Act” was erroneous, Catawba Cnty., 571 F.3d at 41, they have to demonstrate more than mere disagreement with the EPA’s reasoning. Barring an unreasonable or irrational application of the “scientific data within [the EPA’s] technical expertise,” City of Waukesha, 320 F.3d at 247, we cannot say that the EPA acted arbitrarily or capriciously. The record plainly shows that the EPA “considered all relevant factors and articulated a ‘rational connection between the facts found and the choice made’” when it declined to accept the Texas Commission’s source-apportionment model without modification. Catawba Cnty., 571 F.3d at 41 (quoting Burlington Truck Lines, 371 U.S. at 168, 83 S.Ct. 239). We therefore hold that the EPA did not act arbitrarily or capriciously when it did so.

3. The Constitutional Challenges

In this section, we address three constitutional challenges that Texas, Wise County, and the Texas Commission on Environmental Quality (collectively, Texas State Petitioners) raise to the EPA’s designation of Wise County, Texas as a nonattainment area.

i. The Tenth Amendment & The Spending Clause

The Texas State Petitioners, joined by the Mississippi Petitioners, argue that § 7407(d)(1)(B) and related sections of the Clean Air Act — at least to the extent that they authorize the EPA to override the State’s designation and declare Wise County a nonattainment area — violate the Tenth Amendment and exceed the Congress’ authority under the Spending Clause.

First, the Texas State Petitioners maintain that § 7407(d)(1)(B) unlawfully permits the EPA to “commandeer[ ] State regulators to enforce a federal regulatory program.” State & County Br. 32. The section grants the EPA authority to “make such modifications as the Administrator deems necessary to the designations of the areas ... submitted [by the States].” 42 U.S.C. § 7407(d)(l)(B)(ii). According to the petitioners, “[w]hen EPA overrides a State, it compels State regulators to en*175force a myriad of federal requirements involving emissions controls, clean fuel programs, transportation and land use limitations in the designated area.” State & County Br. 33 (citing 42 U.S.C. §§ 7511 et seq. (outlining requirements specific to ozone nonattainment areas)).

The Texas State Petitioners are correct that “the Federal Government may not compel the States to implement ... federal regulatory programs,” Printz v. United States, 521 U.S. 898, 925, 117 S.Ct. 2365, 138 L.Ed.2d 914 (1997).20 But the Clean Air Act does not do that. Instead, the statutory scheme authorizes the EPA to promulgate and administer a federal implementation plan of its own if the State fails to submit an adequate state implementation plan. See 42 U.S.C. § 7410(c). And as we recently noted, the Supreme Court has “repeatedly affirm[ed] the constitutionality of federal .statutes that allow States to administer federal programs but provide for direct federal administration if a State chooses not to administer it.” Texas v. EPA, 726 F.3d 180, 196-97 (D.C.Cir.2013) (citing New York v. United States, 505 U.S. 144, 167-68, 173-74, 112 S.Ct. 2408, 120 L.Ed.2d 120 (1992); Hodel v. Va. Surface Mining & Reclamation Ass’n, Inc., 452 U.S. 264, 288, 101 S.Ct. 2352, 69 L.Ed.2d 1 (1981)). Here, too, the “full regulatory burden will be borne by the Federal Government” if a State chooses not to submit an implementation plan. Va. Surface Mining & Reclamation Ass’n, 452 U.S. at 288, 101 S.Ct. 2389. Under these circumstances, “there can be no suggestion that the Act commandeers ... the States.” Id.

Second, the Texas State Petitioners maintain that the Clean Air Act’s sanctions for noncompliant states impose such a steep price that State officials effectively have no choice but to comply — in contravention of the Supreme Court’s decision in National Federation of Independent Business v. Sebelius (NFIB), — U.S. —, 132 S.Ct. 2566, 2603, 183 L.Ed.2d 450 (2012) (plurality opinion). See State & County Br. 33-34. The Act requires the EPA to impose sanctions on a State that fails to submit an adequate plan or implement an approved plan if it does not correct the deficiency within 18 months. See 42 U.S.C. § 7509(a). The focus of the petitioners’ challenge is the sanction regarding federal highway funds. Under the Act, the EPA Administrator may prohibit the approval of any transportation projects or grants within the nonattainment area, except those that the Secretary of Transportation determines are intended to resolve a demonstrated safety problem and will likely result in a reduction in accidents. Id. § 7509(b)(1)(A). The Secretary of Transportation may also continue to approve a number of other kinds of projects and grants, notwithstanding the EPA Administrator’s prohibition. Id. § 7509(b)(l)(B)(i)-(viii) (authorizing continued approval of projects and grants including capital programs for public transit, projects affecting bus lanes and high occupancy vehicle lanes, programs that improve traffic flow, and programs that “would improve air quality and would not encourage single occupancy vehicle capacity”).

*176As Chief Justice Roberts noted in NFIB, the Supreme Court has “long recognized that Congress may use” the power given it by the Spending Clause “to grant federal funds to the States] and may condition such a grant upon the States’ ‘taking certain actions that Congress could not require them to take.’ ” NFIB, 132 S.Ct. at 2601 (quoting Coll. Sav. Bank v. Fla. Prepaid Postsecondary Educ. Expense Bd., 527 U.S. 666, 686, 119 S.Ct. 2219, 144 L.Ed.2d 605 (1999)). “Such measures ‘encourage a State to regulate in a particular way, [and] influenc[e] a State’s policy choices.’ ” Id. at 2601-02 (quoting New York, 505 U.S. at 166, 112 S.Ct. 2408) (alterations in original). “The conditions imposed by Congress ensure that the funds are used by the States to ‘provide for the ... general Welfare’ in the manner Congress intended.” Id. at 2602 (quoting U.S. Const., art. I, § 8, cl. 1).

“At the same time,” the Chief Justice continued, the Court’s “cases have recognized limits on Congress’s power under the Spending Clause to secure state compliance with federal objectives.” Id. The Court has “repeatedly characterized ... Spending Clause legislation as ‘much in the nature of a contract.’ ” Id. (quoting Barnes v. Gorman, 536 U.S. 181, 186, 122 S.Ct. 2097, 153 L.Ed.2d 230 (2002) (quoting Pennhurst State Sch. & Hosp. v. Halderman, 451 U.S. 1, 17, 101 S.Ct. 1531, 67 L.Ed.2d 694 (1981))). “The legitimacy of Congress’s exercise of the spending power ‘thus rests on whether the State voluntarily and knowingly accepts the terms of the contract.’ ” Id. (quoting Pennhurst, 451 U.S. at 17, 101 S.Ct. 1531) (some internal quotation marks omitted). “Congress may use its spending power to create incentives for States to act in accordance with federal policies,” the Chief Justice concluded, “[b]ut when ‘pressure turns into compulsion,’ the legislation runs contrary to our system of federalism.” Id. (quoting Steward Mach. Co. v. Davis, 301 U.S. 548, 590, 57 S.Ct. 883, 81 L.Ed. 1279 (1937)).21

In NFIB, the Court struck down — as in excess of the Congress’ authority under the Spending Clause — a provision of the Affordable Care Act (ACA) that expanded the scope of the Medicaid program arid increased the number of individuals the States had to cover. Although the Act increased federal funding to cover much of the States’ costs in expanding Medicaid coverage, it also provided that, if a State did not comply with the Act’s new coverage requirements, it could lose not only the new federal funding, but all of its existing federal Medicaid funds. NFIB, 132 S.Ct. at 2582. The Chief Justice’s plurality opinion — for himself and Justices Breyer and Kagan' — -controls our decision on this issue.22

*177In addressing the question of overbearing financial coercion, the Chief Justice first discussed Dole, 483 U.S. 203, 107 S.Ct. 2793, in which the Court rejected such a challenge. In that case, the Congress had threatened to withhold 5 per cent of a State’s federal highway funding unless the State raised its drinking age to 21. The Chief Justice noted that, although “the condition was ‘directly related to one of the main purposes for which highway funds are expended — safe interstate travel,’ ” it “was not a restriction on how the highway funds — set aside for specific highway improvement and maintenance efforts — were to be used.” NFIB, 132 S.Ct. at 2604 (quoting Dole, 483 U.S. at 208, 107 S.Ct. 2793). “[Ajccordingly,” he said, the Dole Court “asked whether ‘the financial inducement offered by Congress’ was ‘so coercive as to pass the point at which pressure turns into compulsion.’ ” Id. (quoting Dole, 483 U.S. at 211, 107 S.Ct. 2793) (some internal quotation marks omitted). The Court answered that this monetary sanction was not impermissibly coercive, but rather offered only “relatively mild encouragement to the states” because “all South Dakota would lose if she adheres to her chosen course as to a suitable minimum drinking age is 5%” of her federal highway funds. Dole, 483 U.S. at 211, 107 S.Ct. 2793; see NFIB, 132 S.Ct. at 2604. “In fact,” as the Chief Justice further noted in NFIB, “the federal funds at stake constituted less than half of one percent of South Dakota’s budget at the time.” NFIB, 132 S.Ct. at 2604.

In NFIB, the Chief Justice found that, as in Dole, the conditions the ACA imposed on the States did not “govern the use of’ the new funds it granted to the States, but rather took “the form of threats to terminate other significant independent grants” already in existence. Id. Accordingly, he said, “the conditions are properly viewed as a means of pressuring the States to accept policy changes” and their level of coerciveness therefore had to be evaluated. Id. Upon doing so, the Chief Justice found the ACA’s financial sanction to be “a gun to the head,” in contrast to the “mild encouragement” in Dole. Id. A State that opted out of the ACA’s Medicaid expansion stood “to lose not merely ‘a relatively small percentage’ of its existing Medicaid funding, but all of it.” Id. (quoting Dole, 483 U.S. at 211, 107 S.Ct. 2793). That, the Chief Justice found, could amount to “over 10 percent of a State’s overall budget.” Id. at 2604-05.

In the case now before us, the Congress has conditioned some federal highway funding on Texas’s adoption of an adequate implementation plan. This condition, like the one at issue in Dole, is — at least arguably — not a restriction on how the highway funds are to be used, but rather an incentive to encourage States to take action in a related policy area. But see discussion infra. Although as discussed below we are uncertain whether that alone is sufficient to trigger a coer-civeness inquiry, we will proceed to evaluate the coercive effect of section 7509(b). For the following reasons, we find that the potential funding sanctions contained in section 7509(b) of the Clean Air Act are hot nearly as coercive as those in the ACA.

First, unlike the situation in NFIB and like that in Dole, a noncompliant State does not risk losing all federal funding for an existing program. To the contrary, the EPA Administrator can only prohibit funding for transportation projects or grants applicable to the nonattainment area. 42 U.S.C. § 7509(b)(1)(A); 40 C.F.R. § 52.31(b)(3), (e)(2) (providing that the “highway funding sanction shall apply ... only to ... areas that are designated non-attainment”); see Virginia v. Browner, 80 *178F.3d 869, 881 (4th Cir.1996) (“[A] state does not lose any highway funds that would be spent in areas of the state that are in attainment.”)- Even within the non-attainment area, the Administrator may not prohibit the approval of projects or grants that the Secretary of Transportation determines are intended to resolve a demonstrated safety problem and will likely result in a reduction in accidents. 42 U.S.C. § 7509(b)(1)(A). Indeed, the Secretary of Transportation may continue to approve a number of other kinds of projects and grants as well, including those that “would improve air quality.” Id. § 7509(b)(1)(B)(viii); see id. § 7509(b)(l)(B)(i)-(viii).

Second, the threatened loss of federal highway funding does not even approach the “over 10 percent of a State’s overall budget” at issue in NFIB. Texas advises us that it received more than $3 billion in federal highway and transit funds in 2013. State & County Br. 33 n. 29. Even if all of that were withheld, it would still have amounted to less than 4 per cent of the State’s 2013 budget.23 But as noted above, Texas does not stand to lose all of its highway funds. The potential sanction applies, at most, to highway funds for projects in nonattainment areas. Wise County is the only county for which the petitioners make a Tenth Amendment argument, and because it is only one of 254 Texas counties, it is unlikely that the loss of even all of that county’s federal highway funds would put a serious dent in the State’s total budget.24 Moreover, as also noted above, it is unlikely that even that one county would lose all of its federal highway funding because the potential sanction does not extend to funding for a list of enumerated projects. See 42 U.S.C. § 7509(b)(1)(A), (B)(i)-(viii).

In short, it is clear that Texas does not risk losing anywhere near the percentage of its federal funding — either for the program at issue or of its overall budget — that the Court found fatal in NFIB. Precisely how much less, we do not know. But the burden of establishing unconstitutionality is on the challenger, and Texas has failed to provide the necessary information. That failure is further ground for rejecting the State’s constitutional challenge. See NFIB, 132 S.Ct. at 2662 (joint opinion of Scalia, Kennedy, Thomas, and Alito, JJ.) (“[CJourts should not conclude that legislation is unconstitutional on this ground unless the coercive nature of an offer is unmistakably clear.”); see also United States v. Morrison, 529 U.S. 598, 607, 120 S.Ct. 1740, 146 L.Ed.2d 658 (2000) (requiring a “plain showing” of unconstitutionality); United States v. Bland, 472 F.2d 1329, 1334 (D.C.Cir.1972) (en banc > (noting that “the burden of establishing the unconstitutionality of a statute rests on him who assails it”).

Finally, although we have concluded that the highway sanction is not unconstitutionally coercive, we note some uncertainty as to whether a coerciveness inquiry was re*179quired. There are two circumstances that may distinguish this case from those in which the Supreme Court has found such an inquiry necessary.

First, as described in NFIB, the inquiry in Dole was triggered by the fact that the Congress had imposed a condition that did not restrict how the federal highway funds at issue were to be used. Here, by contrast, the condition and sanction do redirect the federal highway funds of noncomplying states to programs of the Congress’ choosing, including those that “would improve air quality and would not encourage single occupancy vehicle capacity.” 42 U.S.C. § 7509(b)(l)(B)(viii); see id. § 7509(b)(l)(B)(i)-(viii). As the Senate Committee Report on the 1990 Clean Air Act amendments explains, for nonattainment areas in States that fail to submit an adequate SIP, “Federal transportation investments” are “shifted to transportation programs that are designed to provide alternatives to the single occupancy vehicle and that contribute to reducing future [vehicle miles traveled].” S.Rep. No. 101-228, at 26 (1989).

Second, the condition at issue in Dole — • which required the States to raise their drinking age to 21 — was also, at the time of South Dakota’s challenge, a new condition that had not been part of the original program. In NFIB, although the condition was a restriction on how Medicaid funds could be spent, Chief Justice Roberts found that the condition was also a new one. “Indeed,” he stressed, “the manner in which the expansion is structured indicates that while Congress may have styled the expansion a mere alteration of existing Medicaid, it recognized it was enlisting the States in a new health care program.” NFIB, 132 S.Ct. at 2606. This was important, he said, because “Spending Clause legislation [is] much in the nature of a contract,” id. at 2602 (internal quotation marks omitted), and “[t]hough Congress’ power to legislate under the spending power is broad, it does not include surprising participating States with post-acceptance or retroactive conditions,” id. at 2606 (internal quotation marks omitted). In both Dole and NFIB, the condition at issue was “new” in two senses of the word: Both conditions had been recently enacted at the time of the litigation, and both conditions imposed additional requirements with which States had to comply to continue receiving preexisting federal funding.

Neither the Clean Air Act’s requirement to submit an implementation plan, nor its highway funds sanction, is a condition that has been newly imposed on the States. Although both were new in 1977, see Clean Air Act Amendments of 1977, Pub.L. No. 95-95, §§ 103, 176, 91 Stat. 685, 687-88, 749-50 (1977), since then Texas has submitted implementation plans and accepted billions of dollars in highway funding. Accordingly, when the EPA issued the Wise County nonattainment designation in 2012, Texas was not suddenly surprised by dramatically new conditions retroactively imposed after a long period in which the State had accepted and relied upon unconditional federal funding- — as was the case in NFIB.

These differences from the Supreme Court’s precedents create some uncertainty as to whether the coerciveness inquiry employed in Dole and NFIB was even triggered by the Clean Air Act provisions at issue here. Even if it were, the fact that the State has long accepted billions of dollars notwithstanding the challenged conditions may be an additional relevant factor in the contract-like analysis the Court has in mind for assessing the constitutionality of Spending Clause legislation. But we need not resolve that uncertainty today. Because the challenged provisions of the Clean Air Act survive a coerciveness *180inquiry in any event, we reject the Texas' State Petitioners’ challenge to their constitutionality.

ii. The Commerce Clause

The Texas State Petitioners also argue that the Wise County designation exceeds the scope of the Congress’ authority under the Commerce Clause. As explained above, supra § III.F.l, the designation declared that Wise County contributed enough ozone emissions to nearby violations of the NAAQS to warrant its own nonattainment designation. By virtue of that designation, sources of emissions within the county must comply with a variety of additional requirements. See, e.g., 42 U.S.C. § 7502(c)(1) (requiring the implementation of “all reasonably available control measures”); id. § 7502(c)(5) (requiring “permits for the construction and operation of new or modified major stationary sources anywhere in the nonattainment area”).

The Commerce Clause grants the Congress the power “[t]o regulate Commerce ... among the several States.” U.S. Const., art. I, § 8, cl. 3. The Supreme Court has “recognized ... that ‘[t]he power of Congress over interstate commerce is not confined to the regulation of commerce among the states,’ but extends to activities that ‘have a substantial effect on interstate commerce.’ ” NFIB, 132 S.Ct. at 2585-86 (opinion of Roberts, C.J.) (quoting United States v. Darby, 312 U.S. 100, 118-19, 61 S.Ct. 451, 85 L.Ed. 609 (1941)); see United States v. Lopez, 514 U.S. 549, 558-59, 115 S.Ct. 1624, 131 L.Ed.2d 626 (1995). “Congress’s power, moreover, is not limited to regulation of an activity that by itself substantially affects interstate commerce, but also extends to activities that do so only when aggregated with similar activities of others.” NFIB, 132 S.Ct. at 2586 (opinion of Roberts, C.J.) (citing Wickard v. Filburn, 317 U.S. 111, 127-28, 63 S.Ct. 82, 87 L.Ed. 122 (1942)). The question for a court is whether there was a “rational basis” for the Congress’ conclusion that a regulated activity substantially affects interstate commerce. Hodel v. Indiana, 452 U.S. 314, 323-24, 101 S.Ct. 2376, 69 L.Ed.2d 40 (1981); see Nat’l Ass’n of Home Builders v. Babbitt (NAHB), 130 F.3d 1041, 1051 (D.C.Cir.1997) (opinion of Wald, J.)-.

The Texas State Petitioners’ first contention is that the NOx emissions produced by oil and gas activity in the Barnett Shale in Wise County do not “ ‘substantially affect’ interstate commerce,” principally because the emissions are “wholly intrastate.” State & County Br. 36. That premise is unsupported by any proffered evidence and is factually incorrect. The phenomenon of interstate transport of ozone has been thoroughly studied,.and it has been recognized by the Congress, the EPA, the Supreme Court, and this Court.25 The “winds, of course, recognize no [state] boundaries.” United States v. Ford Motor Co., 814 F.2d 1099, 1102 (6th Cir.1987).

*181But even' if the particular emissions from the Barnett Shale stopped at the Texas state line, the regulation of their sources would still be permissible under the Commerce Clause for two reasons. First, “where a general regulatory-statute bears a substantial relation to commerce, the de minimis character of individual instances arising under that statute is of no consequence.” Lopez, 514 U.S. at 558, 115 S.Ct. 1624 (internal quotation marks omitted) (emphasis omitted); see Gonzales v. Raich, 545 U.S. 1, 17, 125 S.Ct. 2195, 162 L.Ed.2d 1 (2005); NAHB, 130 F.3d at 1046 (opinion of Wald, J.). And there is no doubt that the general regulatory scheme of the Clean Air Act has a substantial relation to interstate commerce. Indeed, the same is true even if we focus only upon the Act’s generally applicable ozone provisions.

Moreover, we can find a substantial effect not only by examining the emissions that are produced, but also by examining the activities that the challenged statute regulates to reduce the production of those emissions. See Rancho Viejo, LLC v. Norton, 323 F.3d 1062, 1067 (D.C.Cir.2003); NAHB, 130 F.3d at 1046 & n. 3 (opinion of Wald, J.); id. at 1058 (Henderson, J., concurring). As we explained in Rancho Viejo, on this rationale we “focus[ ] on the activity that the federal government seeks to regulate.” 323 F.3d at 1069; see Morrison, 529 U.S. at 609, 120 S.Ct. 1740 (instructing that “the proper inquiry” is whether the challenge is to “a regulation of activity that substantially affects interstate commerce”) (emphasis added); Lopez, 514 U.S. at 558-59, 115 S.Ct. 1624 (“Congress’ commerce authority includes the power to regulate ... those activities that substantially affect interstate commerce.”) (emphasis added). In Rancho Viejo, we upheld the constitutionality of the Fish and Wildlife Service’s decision to protect an endangered toad species by regulating a housing development, on the ground that the regulated activity, a “202-acre project, located near a major interstate highway, [was] ... presumably being constructed using materials and people from outside the state.” 323 F.3d at 1069 (internal quotation marks omitted). Likewise, in NAHB, we upheld the Service’s decision to protect an endangered fly species by regulating the construction plan for a hospital, on the ground that the commercial land development at issue “ha[d] a plain and substantial effect on interstate commerce.” 130 F.3d at 1059 (Henderson, J., concurring); see id. at 1056 (opinion of Wald, J.).

Here, the activities that the EPA seeks to regulate are the commercial, industrial, and extraction processes that produce the emissions at issue. -See 42 U.S.C. § 7511a; 2008 Designations Rule, 77 Fed.Reg. at 30,089. The nonattainment designation triggers regulatory controls on the sources of those emissions, many of which are indisputably entities engaged in substantial interstate commerce. In the case of Wise County in particular, those entities include multinational companies engaged in the production and sale of oil and gas from the Barnett Shale, including several of the Industrial Petitioners here.26 The restrictions triggered by the nonattainment designation thus affect the conditions under which interstate commerce in oil and gas *182may proceed. And as such, the designation process “regulates and substantially affects commercial ... activity which is plainly interstate.” NAHB, 130 F.3d at 1058 (Henderson, J., concurring).

The Texas State Petitioners’ second contention is that, “[e]ven if incidental emissions do ‘substantially affect’ interstate commerce, they are not ‘quintessentially economic activity’ ” and cannot be regulated under the Commerce Clause. State & County Br. 36. This contention is based on the Court’s- decision in Lopez, which held the Gun-Free School Zones Act unconstitutional in part because the statutory provision at issue, which criminalized the possession of a gun in a school zone,' had “nothing to do with ‘commerce’ or any sort of economic enterprise, however broadly one might define those terms.” Lopez, 514 U.S. at 560-61, 115 S.Ct. 1624; see also Morrison, 529 U.S. at 610-11, 613, 120 S.Ct. 1740. There are two answers to this contention.

First, ozone pollution itself has economic consequences for interstate commerce. The Congress so found in the course of amending the Clean Air Act. See S. Rep. No. 101-228, at 8 (1989) (noting that exposure to ah* pollution costs the United States $40 billion annually in additional health care costs, and documenting, health effects of ozone and other pollutants); id. (noting that “ozone causes annual crop losses of $2 to $3 billion per year”). Although we are not bound by congressional findings, they may assist us in “evaluating] the legislative judgment that the activity in question substantially affected interstate commerce.” Lopez, 514 U.S. at 562-63, 115 S.Ct. 1624; see Rancho Viejo, 323 F.3d at 1069. Indeed, we have previously credited the Congress’ findings regarding ozone pollution, concluding that the Act’s “legislative history and EPA’s report to Congress substantiate the heavy impact ozone pollution has on national health care costs and national agricultural production.” Allied Local, 215 F.3d at 83.

Second, the activities that are ultimately régulated by the designation process are not the ozone precursor “emissions,” but rather the activities that produce the emissions. Those include the operation of power plants, gas processors, and vehicles that produce the emissions. See 42 U.S.C. § 7511a. As we explained in Rancho Vie-jo, the regulated activity in that case was a company’s “planned commercial development, not the arroyo toad that it threatened].” 323 F.3d at 1072. The same point is true here. Just as the Endangered Species Act “does not purport to tell toads what they may or may not do,” id., the Clean Air Act does not tell NOx or VOCs what to do. Rather, it tells the commercial and industrial sources that produce those compounds what they may do.

As we noted in Allied Local, the Supreme Court has long made clear that “ ‘the power conferred by the Commerce Clause [is] broad enough to permit congressional regulation of activities causing air or water pollution, or other environmental hazards that may have effects in more than one State.’ ” Allied Local, 215 F.3d at 83 (quoting Va. Surface Mining & Reclamation Ass’n, 452 U.S. at 282, 101 S.Ct. 2352) (emphasis added); id. (noting that the Supreme Court cited Virginia Surface Mining and Reclamation Association with approval in both Lopez and Morrison). “[B]eeause we are required to accord congressional legislation a ‘presumption of constitutionality,’ ” Rancho Viejo, 323 F.3d at 1069 (quoting Morrison, 529 U.S. at 607, 120 S.Ct. 1740), the petitioners’ inability to establish that emissions-produeing sources in the State do not substantially affect interstate commerce “is fatal to [their] cause,” id.; see Morrison, 529 U.S. at 607, 120 S.Ct. 1740 (“Due respect for the decisions of a coordinate *183branch of Government demands that we invalidate a congressional enactment only upon a plain showing that Congress has exceeded its constitutional bounds.”). The regulation of the sources of Wise County emissions through the Clean Air Act’s designation process lies well within the Congress’ authority to regulate interstate commerce.

Hi. The Due Process Clause

The Texas State Petitioners’ third constitutional challenge maintains that the EPA’s designation of Wise County violated the Due Process Clause because the former Administrator of EPA Region 6, A1 Armendariz, failed to disqualify himself from the proceedings.

According to the petitioners, Armendar-iz should have disqualified himself for four reasons. First, Armendariz has a history of working for environmental advocacy groups. Second, a report he authored as an advocate before joining the EPA concluded that emissions from the Barnett Shale were contributing significantly to local and global pollution. Third, a speech Armendariz gave after joining the EPA analogized his aggressive enforcement policy against oil and gas companies that “are not complying with the law” to the way “Romans used to conquer those villages in the Mediterranean” by “crucify[ing]” the first people they saw. Terrence Henry, Texas EPA Official Apologizes for ‘Crucify Them’ Comments, Apr. 26, 2012, State Impact NPR, http://stateimpact.npr.org/ texas/2012/04/26/epa-offieial-apologizes-for-crucify-comments (quoting Armendariz). “You make examples out of people who ... are not complying with the law,” Ar-mendariz said. “There’s a deterrent factor.... And they decide at that point that it’s time to clean up.” Id27 Finally, in the petitioners’ view, “[njormally, the prevailing wind direction and EPA-standard modeling would have led EPA to accept” Texas’s designation of Wise County as attainment. State & County Br. 38. All of this, the petitioners argue, “create[s] a presumption that the Agency’s mind was closed and it was unwilling or unable to rationally consider arguments against non-attainment.” Id. at 37.

In Air Transport Association of America, Inc. v. National Mediation Board, 663 F.3d 476 (D.C.Cir.2011), we repeated this circuit’s approach to the kind of claim that the petitioners raise here. “Decisionmak-ers violate the Due Process Clause and must be disqualified,” we said, “when they act with an ‘unalterably closed mind’ and are ‘unwilling or unable’ to rationally consider arguments.” Id. at 487 (quoting Ass’n of Nat’l Advertisers, Inc. v. FTC, 627 F.2d 1151, 1170, 1174 (D.C.Cir.1979)). “[A]n individual should be disqualified from rule-making only when there has been a clear and convincing showing that the ... member has an unalterably closed mind on matters critical to .the disposition of the proceeding.” Id. (quoting C & W Fish Co., Inc. v. Fox, 931 F.2d 1556, 1564 (D.C.Cir.1991) (internal quotation marks omitted)). The four arguments advanced by the Texas State Petitioners are insufficient to make that “clear and convincing” showing.28

*184Our decision in C & W Fish Company establishes that neither Armendariz’ employment history nor the report he authored before joining the EPA required his disqualification. There, we considered the impartiality of an agency administrator who had previously served as the chairman of a group advocating for the precise agency policy at issue in the case, and who after his appointment remarked that there was “no question” that the policy should be implemented. C & W Fish Co., 931 F.2d at 1564. Those circumstances, we said, did “not even approach a ‘cléar and convincing showing’ that [the administrator] had an ‘unalterably closed mind.’” Id. at 1565.

The petitioners’ third argument is also unpersuasive. There is no doubt that Ar-mendariz’ “crucifixion” comments were offensive. But that does not suffice to make the requisite showing. The comments described Armendariz’ general approach to enforcement, but were neither specifically about the designation process nor specifically targeted at production from the Barnett Shale. Accordingly, they did not reveal Armendariz’ views on “matters critical to the disposition of the proceeding.” Ass’n of Nat’l Advertisers, 627 F.2d at 1170. And even if they had, they would not alone demonstrate an unalterably closed mind on the subject. See C & W Fish Co., 931 F.2d at 1565 (“ ‘We would eviscerate the proper evolution of policy-making were we to disqualify every administrator who has opinions on the correct course of his agency’s future actions.’ ” (quoting Ass’n of Nat’l Advertisers, 627 F.2d at 1174)).

Finally, we cannot infer bias from the fact that, in the opinion of the petitioners, the computer modeling supported an attainment designation for Wise County. As we held in C & W Fish Company, “we reject the suggestion that we look to the adequacy of [an agency official’s] examination of the facts and issues in order to determine whether he was biased.” 931 F.2d at 1564. Rather, “[w]hether [the official] weighed the facts properly is to be examined only in determining if his decision was arbitrary or capricious.” Id. at 1564-65. And that is an examination that we separately undertake in section III.F.2, supra.

For the foregoing reasons, we reject the petitioners’ three constitutional challenges to the designation of Wise County as a nonattainment area.

4. The Remaining Challenges

Finally, the Texas State Petitioners argue that we should vacate the EPA’s Wise County nonattainment designation because the EPA (1) failed to comply with the Information Quality Act, (2) failed to promulgate regulations defining the terms “necessary” and “contribute,” (3) concluded that Wise County emissions “can” contribute to NAAQS violations when it was statutorily required to conclude that Wise County “did” contribute, and (4) failed to give them “fair notice” of the' EPA’s requirements. State & County Br. 46-52. We reject all four contentions.

First, the Texas State Petitioners urge us to conclude that the Information Quality Act requires the EPA to use “the best available science and supporting studies conducted in accordance with sound and objective scientific practices” in making NAAQS designations, State & County Br. 46 (citing Prime Time Int’l Co. v. Vilsack, 599 F.3d 678, 685-86 (D.C.Cir.2010)), and that the EPA failed to do so here. But almost every court that has addressed an Information Quality Act challenge has held that the statute “creates no legal rights in any third parties,” Salt Inst. v. Leavitt, 440 F.3d 156, 159 (4th Cir.2006);29 see also Harkonen v. U.S. *185Dep’t of Justice, No. C 12-629 CW, 2012 WL 6019571, at *11 (N.D.Cal. Dec. 3, 2012) (collecting cases). And this Court has held that the Information Quality Act is not “an independent measure of EPA’s NAAQS decision.” Mississippi, 744 F.3d at 1347. The purpose of the Information Quality Act is to “ensur[e] and maximizefe] the quality, objectivity, utility, and integrity of information (including statistical information) disseminated by Federal agencies” and does not constitute a statutory mechanism by which the EPA’s conclusions reached while making its nonattainment determinations can be challenged. See 44 U.S.C. § 3516 note (emphasis added).

Second, the Texas State Petitioners argue that the EPA should define the terms “contribute” and “necessary” through administrative rulemaking in order to rein in the “boundless override discretion” it uses to “commandeer[ ]” states to “enforce its massive regulatory scheme.” See State & County Br. 48. Our Catawba County holding forecloses this argument. There, we held that the EPA was “free to adopt a totality-of-the-circumstances test to implement a statute that confers broad discretionary authority.” Catawba Cnty., 571 F.3d at 39. Finally, the Texas State Petitioners offer no reason why the word “necessary,” which the EPA reasonably interpreted as authorizing modification of a state’s recommended designation that does “not meet the statutory requirements or [was] otherwise inconsistent with the facts or analysis deemed appropriate by the EPA,” see 2008 Designations Rule, 77 Fed. Reg. at 30,090, must be defined via rule-making.

Third, the Texas State Petitioners argue that the EPA exceeded its authority under the Clean Air Act because it concluded that Wise County emissions “can” contribute to NAAQS violations, whereas the Act authorizes a finding that Wise County “does” so contribute. See State & County Br. 50. This argument is premised on the EPA’s response to a petition for reconsideration challenging the Wise County nonattainment designation, to which the EPA responded that “the Wise County -emissions are large enough that they can contribute to ozone exceedances on certain days.” EPA Response to Pet. for Reconsideration from Wise Cnty., Office of the Cnty Judge at 2 (emphasis added). But read in toto, the EPA’s justification for including Wise County in the Dallas-Fort Worth nonattainment area-was anything but theoretical:

Wise County [h]as 2008 NEI emissions of 11,911 tons of NOx and 17,609 tons of VOC; there are 60 people per square mile; has a 2010 population of 59,127 with a growth rate of 5.9 percent between 2000 and 2010; total VMT is 969 million. The close proximity of these comparatively high emissions to violating monitors indicates that this county should be included in the nonattainment area. The high growth in these emissions is due in large part to growth in emissions from Barnett Shale gas production development, but also due to growth in population. Examination of back trajectories indicates that at times emissions from Wise County contribute to observed violations in the area and also to observed violations that have helped set the DFW area DV in the past. Source apportionment modeling for a portion of an ozone season indicates that emissions from Wise County can contribute to observed violations in the DFW nonattainment area. These factors support the inclusion of Wise County in the nonattainment area.

*186Final Dallas-Fort Worth Area Designations at 23. Read in context, we conclude that the EPA in fact found that Wise County does contribute to NAAQS violations in the Dallas-Fort Worth area.

Fourth, the Texas State Petitioners argue the EPA failed to provide them with “fair notice” of its requirements. Even assuming the fair notice doctrine applies, cf. Ark. Dep’t of Human Servs. v. Sebelius, 818 F.Supp.2d 107, 120-21 (D.D.C.2011), the EPA did not violate it. The fair notice doctrine, which is couched in terms of due process, provides redress only if an agency’s interpretation is “so far from a reasonable person’s understanding of the regulations that they could not have fairly informed the regulated party of the agency’s perspective.” United States v. Chrysler Corp., 158 F.3d 1350, 1354 (D.C.Cir.1998) (alteration omitted). Here, the EPA not only provided the 2008 Guidance to aid the states in making their initial designations, it also provided a preliminary technical support document to each state before finalizing any of its proposed modifications to the state’s initial designations. See, e.g., Preliminary Dallas-Fort Worth Area Designations. The technical support document, in turn, gave each state a precise explication of all proposed EPA modifications as a roadmap to use during the 120-day comment period. Simply put, the EPA set forth its analysis, provided an opportunity to rebut its conclusions and ultimately explained why it had not changed its mind. Accordingly, the Texas State Petitioners’ fair notice doctrine argument is meritless.

For the foregoing reasons, the consolidated petitions for review are denied.

So ordered.

PER CURIAM:

Petitioners challenge EPA’s new source performance standards for nitrogen oxides emissions from utility and industrial boilers. We conclude that EPA did. not exceed its discretion under section 111 of the Clean Air Act in promulgating these standards, and therefore deny the petitions.

Fossil-fuel fired steam generating units (“boilers”) emit nitrogen oxides (N0X), air pollutants that can cause deleterious health effects and contribute to the formation of acid rain. Section 111 of the Clean Air Act requires EPA to establish performance standards for the emission of N0X from newly constructed boilers; these “new source performance standards” are to be set at a level that

reflects the degree of emission limitation achievable through the application of the best system of emission reduction which (taking into account the cost of achieving such reduction and any nonair quality health and environmental impact and energy requirements) the Administrator determines has been adequately demonstrated.

42 U.S.C. § 7411(a)(1). In its 1990 Clean Air Act Amendments Congress specifically directed EPA to exercise its section 111 authority and establish new NOx standards that incorporate “improvements in methods for the reduction of emissions of oxides of nitrogen.” 42 U.S.C. § 7651f(c)(1).

In response to these statutory mandates, EPA promulgated a rule lowering its NOx new source performance standards to .15 lb/MMBtu (pounds of NOx emitted per million BTU burned) for utility boilers1 and .20 lb/MMBtu for industrial boilers. See 63 Fed.Reg. 49,442, 49,443 (1998) (to be codified at 40 C.F.R. pt. 60). These standards reflect the level of NOx emissions achievable by what EPA considers to be the “best demonstrated system” of emissions reduction: the use of selective catalytic reduction (SCR) in combination with combustion control technologies.2 *933Petitioners’ central claim is that EPA selected SCR as the basis for its NOx standards without properly balancing the factors that section 111 requires it to “take into account.” Because section 111 does not set forth the weight that be should assigned to each of these factors, we have granted the agency a great degree of discretion in balancing them, see, e.g., New York v. Reilly, 969 F.2d 1147, 1150 (D.C.Cir.1992); EPA’s choice will be sustained unless the environmental or economic costs of using the technology are exorbitant. See National Asphalt Pavement Ass’n v. Train, 539 F.2d 775, 786 (D.C.Cir.1976).

Petitioners argue that SCR is not the “best demonstrated system” under section 111 because the incremental cost of reducing NOx emissions is considerably higher with SCR than with combustion controls. Recent improvements in combustion controls will enable many boilers to attain emissions levels close to EPA’s SCR-based standards; accordingly, petitioners assert that EPA should have based its standards on these less expensive technologies. However, in light of EPA’s unchallenged findings showing that the new standards will only modestly increase the cost of producing electricity in newly constructed boilers, see 62 Fed.Reg. 36,948, 36,958 (1997) (proposed NOx revisions), we do not think that EPA exceeded its considerable discretion under section 111. Moreover, petitioners’ argument stressing the comparable environmental merits of advanced combustion controls is to a certain extent self-defeating, since the new source performance standards set by EPA are not technology-forcing, and continuing advances in combustion control technologies will reduce the amount of NOx reduction that must be captured by the more expensive SCR technology.

It was also within EPA’s discretion to issue uniform standards for all utility boilers, rather than adhering to its past practice of setting a range of standards based on boiler and fuel type. See, e.g., 44 Fed.Reg. 33,580 (1979) (establishing varying NOx emissions standards for utility boilers). Petitioners recognize that EPA is not required by law to subcategorize— section 111 merely states that “the Administrator may distinguish among classes, types, and sizes within categories of new sources,” 42 U.S.C. § 7411(b)(2) (emphasis added) — but argue that it was arbitrary and capricious -for EPA to decline to do so. EPA explains that its change to uniform standards is justified by SCR’s performance characteristics: Unlike the technologies on which past new source performance standards were based, flue gas treatment technologies like SCR limit NOx emissions after combustion, and the effectiveness of SCR is thus far less dependent upon boiler design or fuel type. Petitioners respond that there are reasons to expect SCR to perform less adequately on boilers burning high-sulfur coals, but EPA collected continuous emissions monitoring data on two high-sulfur coal-fired utility boilers that showed that the .15 lb/MMBtu standard was achievable, and supplemented this study with similar evidence from foreign utility boilers. EPA also considered petitioners’ concerns about the impact of alkaline metals on the performance of the catalyst used in the SCR process, and concluded that such “catalyst poisoning” is not a significant problem in coal-fired boilers. See 63 Fed.Reg. at 49,445. Mindful of the high degree of deference we must show to EPA’s scientific judgment, see, e.g., Appalachian Power Co. v. EPA, 135 F.3d 791, 801-02 (D.C.Cir.1998), we accept these determinations and sustain EPA’s uniform standard for utility boilers.

Petitioners offer a broader challenge to EPA’s .20 lb/MMBtu standard for industrial boilers, claiming that SCR is not “adequately demonstrated” for any coal-fired industrial boilers. EPA was unable to collect emissions data for the application of SCR to these boilers, but this absence of *934data is not surprising for a new technology like SCR, nor does it in and of itself defeat EPA’s standard. Because it applies only to new sources, we have recognized that section 111 “looks toward what may fairly be projected for the regulated future, rather than the state of the art at present.” Portland Cement Ass’n v. Ruckelshaus, 486 F.2d 375, 391 (D.C.Cir.1973). Of course, where data are unavailable, EPA may not base its determination that a technology is adequately demonstrated or that a standard is achievable on mere specular tion or conjecture, see, e.g., National Asphalt Pavement Ass’n, 539 F.2d at 787, but EPA may compensate for a shortage of data through the use of other qualitative methods, including the reasonable extrapolation of a technology’s performance in other industries. See, e.g., Weyerhaeuser Co. v. Costle, 590 F.2d 1011, 1054 n. 70 (D.C.Cir.1978).

EPA has done precisely that here, concluding from its study of utility boilers that SCR is “adequately demonstrated” and the .20 lb/MMBtu standard is “achievable” for coal-fired industrial boilers as well. Utility and industrial boilers are similar in design and both categories of boilers can attain similar levels of NOx emissions reduction through combustion controls, which means that SCR will be required to capture comparable quantities of NOx for both boiler types. While petitioners argue that SCR is less likely to be effective on industrial boilers because they have widely fluctuating load cycles, EPA has shown that SCR can be successfully applied to coal-fired utility boilers under a “wide range of operating conditions” including those analogous to the load cycles of industrial boilers. 63 Fed.Reg. at 49,-444. We think that it was reasonable for EPA to extrapolate from its studies of utility boilers in setting an SCR-based new source performance standard for coal-fired industrial boilers.3

We also sustain EPA’s application of the .20 lb/MMBtu standard to combination boilers, which simultaneously combust a mixture of fuels. The preexisting NOx emissions standards established a range of values for combustion boilers that varied by fuel type: while combination boilers burning natural gas with non-coal solid fuels (e.g., wood) were subject to a .30 lb/MMBtu standard, the performance standards for combination boilers combusting coal with oil or natural gas were determined based upon the proportion of the boiler’s total heat input provided by each fuel. See 51 Fed.Reg. 42,768, 42,790 (1986). It is difficult to understand petitioners’ objection to the application of the industrial boiler standard to boilers burning natural gas and wood. A reduction of that standard from .30 to .20 lb/MMBtu is perfectly reasonable in light of the significant advances in NOx emissions technology since 1986; indeed, EPA studies show that wood-fired boilers can reach emissions levels far lower than .20 lb/MMBtu through the application of flue gas treatment technologies. And our conclusion that the .20 lb/MMBtu standard is achievable for boilers burning only coal necessarily defeats petitioners’ objection that the industrial boiler standard is unreasonable as applied to combination boilers burning coal simultaneously with other fuels with lower NOx emissions characteristics.

Petitioners’ final objection is to EPA’s valuation of steam energy produced by “cogeneration facilities.” EPA’s adoption of an output-based standard for utility boilers raised the question of how to calculate the energy produced by these units, which generate thermal steam energy in addition to electrical energy. Steam ener*935gy produced by cogeneration facilities is exported for several different industrial uses; however, because of inefficiencies in transporting and converting steam, only a fraction of steam energy produced by co-generation facilities is actually used in the industrial process. EPA resolved this problem by assigning a 50% credit for steam energy when determining a cogen-eration unit’s output. See 63 Fed.Reg. at 49,447. Petitioners describe this credit as an arbitrary and capricious “discounting” of steam energy’s value, but it just as easily could be called a subsidy: The maximum efficiency for the conversion of steam to electrical energy is only 38%, and EPA’s final rule justifies the 50% credit on the ground that it will encourage cogeneration. Id. In light of the difficulties that would attend calculating the useful energy of steam heat produced by cogeneration facilities on a unit-by-unit basis, we conclude that EPA’s resolution of this issue was acceptable.

The petitions for review are denied.

So ordered.

K. K. HALL, Circuit Judge.

Citizens Against the Refinery’s Effects (CARE) appeals from a final ruling by the Administrator of the Environmental Protection Agency (EPA) approving the Virginia State Implementation Plan (SIP) for reducing hydrocarbon pollutants. The plan requires the Virginia Highway Department to decrease usage of a certain type of asphalt, thereby reducing hydrocarbon pollution by more than enough to offset expected pollution from the Hampton Roads Energy Company’s (HREC) proposed refinery. We affirm the action of the administrator in approving the state plan.

The Act

The Clean Air Act establishes National Ambient Air Quality Standards (NAAQS) for five major air pollutants.1 42 U.S.C. § 7409; 40 CFR § 50 (1976). The EPA has divided each state into Air Quality Control Regions (AQCR)2 and monitors each region to assure that the national standard fór each pollutant is met. 42 U.S.C. § 7407. Where the standard has not been attained for a certain pollutant, the state must develop a State Implementation Plan designed to bring the area into attainment within a certain period. 42 U.S.C. § 7410. In addition, no new source of that pollutant may be constructed until the standard is attained. 40 CFR § 51.18 (1973).

The Clean Air Act created a no-growth environment in areas where the clean air requirements had not been attained. EPA recognized the need to develop a program that encouraged attainment of clean air standards without discouraging economic growth. Thus the agency proposed an Interpretive Ruling in 1976 which allowed the states to develop an “offset program” within the State Implementation Plans. 41 Fed. Reg. 55524 (1976). The offset program, later codified by Congress in the 1977 Amendments to the Clean Air Act, permits the states to develop plans which allow construction of new pollution sources where accompanied by a corresponding reduction in an existing pollution source. 42 U.S.C. *185§ 7502(b)(6) and § 7503. In effect, a new emitting facility can be built if an existing pollution source decreases its emissions or ceases operations as long as a positive net air quality benefit occurs.

If the proposed factory will emit carbon monoxide, sulfur dioxide, or particulates, the EPA requires that the offsetting pollution source be within the immediate vicinity of the new plant. The other two pollutants, hydrocarbons and nitrogen oxide, are less “site-specific,” and thus the ruling permits the offsetting source to locate anywhere within a broad vicinity of the new source.3

The offset program has two other important requirements. First, a base time period must be determined in which to calculate how much reduction is needed in existing pollutants to offset the new source. This base period is defined as the first year of the SIP or, where the state has not yet developed a SIP, as the year in which a construction permit application is filed. 41 Fed.Reg. 55529 (1976). Second, the offset program requires that the new source adopt the Lowest Achievable Emissions Rate (LAER) using the most modern technology available in the industry. 41 Fed.Reg. 55528-9 (1976).

The Refinery

HREC proposes to build a petroleum refinery and offloading facility in Portsmouth, Virginia. Portsmouth has been unable to reduce air pollution enough to attain the national standard for one pollutant, photochemical oxidants,4 which is created when hydrocarbons are released into the atmosphere and react with other substances. Since a refinery is a major source of hydrocarbons, the Clean Air Act prevents construction of the HREC plant until the area attains the national standard.

In 1975, HREC applied to the Virginia State Air Pollution Control Board (VSAPCB) for a refinery construction permit. The permit was issued by the VSAPCB on October 8, 1975, extended and reissued on October 5, 1977 after a full public hearing, modified on August 8, 1978, and extended again on September 27, 1979. The VSAPCB, in an effort to help HREC meet the clean air requirements, proposed to use the offset ruling to comply with the Clean Air Act.

On November 28,1977, the VSAPCB submitted a State Implementation Plan to EPA which included the HREC permit The Virginia Board proposed to offset the new HREC hydrocarbon pollution by reducing the amount of cutback asphalt5 used for road paving operations in three highway districts by the Virginia Department of Highways.6 By switching from “cutback” to “emulsified” asphalt, the state can reduce hydrocarbon pollutants by the amount necessary to offset the pollutants from the proposed refinery.

EPA requested some changes in the state plan, including certain monitoring changes and verification from the Virginia Attorney General that the offset program was legally enforceable. The plan was transmitted by the EPA Region III director to EPA headquarters on September 9, 1978. Notices of the proposed plan were published on October 10, 1978 and again on May 1, 1979. 43 Fed.Reg. 46554 (1978). 44 Fed.Reg. 25471 (1979). Numerous comments were received, *186including several from CARE. The EPA administrator carefully considered the comments and .approved the Virginia offset plan on January 31, 1980.

CARE raises four issues regarding the state plan. First, they argue that the geographic area used as the base for the offset was arbitrarily determined and that the area as defined violates the regulations. Second, CARE contends that EPA should have used 1975 instead of 1977 as the base year to compare usage of cutback asphalt. Third, CARE insists that the offset plan should have been disapproved since the state is voluntarily reducing usage of cutback asphalt anyway. Fourth, CARE questions the approval of the plan without definite Lowest Achievable Emissions Rates (LAER) as required by the statute. We reject the CARE challenges to the state plan.

As in Citizens Against Refinery Effects v. EPA, 643 F.2d 178 (4th Cir. 1981), the standard of review here is whether the agency action was arbitrary, capricious, an abuse of discretion, or otherwise not in accordance with law. Citizens to Preserve Overton Park, Inc. v. Volpe, 401 U.S. 402, 91 S.Ct. 814, 28 L.Ed.2d 136 (1971); Appalachian Power Company v. EPA, 477 F.2d 495 (4th Cir. 1973).

The Geographic Area

CARE contends that the state plan should not have been approved by EPA since the three highway-district area where cutback usage will be reduced to offset refinery emissions was artificially developed by the state. The ruling permits a broad area (usually within one AQCR) to be used as the offset basis. 41 Fed.Reg. 55529 (1976).

The ruling does not specify how to determine the area, nor provide a standard procedure for defining the geographic area. 41 Fed.Reg. 55529 (1976). Here the Virginia Board originally proposed to use four highway districts comprising one-half the state as the offset area. When this was found to be much more than necessary to offset pollution expected from the refinery, the state changed it to one highway district plus nine additional counties. Later the proposed plan was again revised to include a geographic area of three highway districts.

The agency action in approving the use of three highway districts was neither arbitrary, capricious, nor outside the statute. First, Congress intended that the states and the EPA be given flexibility in designing and implementing SIPs. See H.R.Rep.No. 95-294, 95th Cong., 1st Sess., reprinted in [1977] U.S.Code Cong. & Ad. News 1077, 1290-92. Such flexibility allows the states to make reasoned choices as to which areas may be used to offset new pollution and how the plan is to be implemented. Second, the offset program was initiated to encourage economic growth in the state. 41 Fed.Reg. 55525 (1976). Thus a state plan designed to reduce highway department pollution in order to attract another industry is a reasonable contribution to economic growth without a corresponding increase in pollution. Third, to be sensibly administered the offset plan had to be divided into districts which could be monitored by the highway department. Use of any areas other than highway districts would be unwieldy and difficult to administer. Fourth, the scientific understanding of ozone pollution is no't advanced to the point where exact air transport may be predicted. Designation of the broad area in which hydrocarbons may be transported is well within the discretion and expertise of the agency.

The Base Year

Asphalt consumption varies greatly from year to year, depending upon weather and road conditions. Yet EPA must accurately determine the volume of hydrocarbon emissions from cutback asphalt. Only then can the agency determine whether the reduction in cutback usage will result in an offset great enough to account for the new refinery pollution. To calculate consumption of a material where it constantly varies, a base year must be selected. In this case, EPA’s Interpretive Ruling establishes *187the base year as the year in which the permit application is made. 41 Fed.Reg. 55529 (1976). EPA decided that 1977 was an acceptable base year. CARE argues that EPA illegally chose 1977 instead of 1975.

Considering all of the circumstances, including the unusually high asphalt consumption in 1977, the selection by EPA of that as the base year was within the discretion of the agency. Since the EPA Interpretive Ruling allowing the offset was not issued until 1976, 1977 was the first year after the offset ruling and the logical base year in which to calculate the offset. Also, the permit issued by the VSAPCB was reissued in 1977 with extensive additions and revisions after a full hearing. Under these circumstances, 1977 appears to be a logical choice of a base year.

The Legally Binding Plan

For several years, Virginia has pursued a policy of shifting from cutback asphalt to the less expensive emulsified asphalt in road-paving operations. The policy was initiated in an effort to save money, and was totally unrelated to a State Implementation Plan.7 Because of this policy, CARE argues that hydrocarbon emissions were decreasing independent of this SIP and therefore are not a proper offset against the refinery. They argue that there is not, in effect, an actual reduction in pollution.

The Virginia voluntary plan is not enforceable and therefore is not in compliance with the 1976 Interpretive Ruling which requires that the offset program be enforceable. 41 Fed.Reg. 55526 (1976). The EPA, in approving the state plan, obtained a letter from the Deputy Attorney General of Virginia in which he stated that the requisites had been satisfied for establishing and enforcing the plan with the Department of Highways. Without such authority, no decrease in asphalt-produced pollution is guaranteed.8 In contrast to the voluntary plan, the offset plan guarantees a reduction in pollution resulting from road-paving operations.

The Lower Achievable Emissions Rate

Finally, CARE argues that the Offset Plan does not provide adequate Lowest Achievable Emission Rates (LAER) as required by the 1976 Interpretive Ruling because the plan contains only a 90% vapor recovery requirement, places an excessive 176.5 ton limitation on hydrocarbon emissions, and does not require specific removal techniques at the terminal. EPA takes the position that the best technique available for marine terminals provides only a 90% recovery and that the 176.5 ton limit may be reduced by the agency after the final product mix at the terminal is determined.

Since the record shows no evidence of arbitrary or capricious action in approving the HREC emissions equipment, the agency determination of these technical matters must be upheld. Udall v. Tallman, 380 U.S. 1, 85 S.Ct. 792, 13 L.Ed.2d 616 (1965).

Conclusion

In approving the state plan, EPA thoroughly examined the data, requested changes in the plan, and approved the plan only after the changes were made. There is no indication that the agency acted in an arbitrary or capricious manner or that it stepped beyond the bounds of the Clean Air Act. We affirm the decision of the administrator in approving the state plan.

*468Justice Ginsburg

delivered the opinion of the Court.

This case concerns the authority of the Environmental Protection Agency (EPA or Agency) to enforce the provisions of the Clean Air Act’s (CAA or Act) Prevention of Significant Deterioration (PSD) program. Under that program, no major air pollutant emitting facility may be constructed unless the facility is equipped with “the best available control technology” (BACT). As added by § 165, 91 Stat. 735, and amended, 42 U. S. C. § 7475(a)(4). BACT, as defined in the CAA, means, for any major air pollutant emitting facility, “an emission limitation based on the maximum degree of [pollutant] reduction . .. which the permitting authority, on a case-by-case basis, taking into account energy, environmental, and economic impacts and other costs, determines is achievable for [the] facility....” § 7479(3).

Regarding oversight, the a struction and one geared specifically to the PSD program. The general prescription, § 113(a)(5) of the Act, authorizes EPA, when it finds that a State is not complying with a CAA requirement governing construction of a pollutant source, to issue an order prohibiting construction, to prescribe an administrative penalty, or to commence a civil action for in-junctive relief. 42 U. S. C. § 7413(a). Directed specifically to the PSD program, CAA §167 instructs EPA to “take such measures, including issuance of an order, or seeking ih-*469junctive relief, as necessary to prevent the construction” of a major pollutant emitting facility that does not conform to the PSD requirements of the Act. 42 U. S. C. § 7477.

In the case before us, “the permitting authority” under §7479(3) is the State of Alaska, acting through Alaska’s Department of Environmental Conservation (ADEC). The question presented is what role EPA has with respect to ADEC’s BACT determinations. Specifically, may EPA act to block construction of a new major pollutant emitting facility permitted by ADEC when EPA finds ADEC’s BACT determination unreasonable in light of the guides §7479(3) prescribes? We hold that the Act confers that checking authority on EPA.

I

A

Congress enacted the Clean Air Amendments of 1970, 84 Stat. 1676, 42 U. S. C. § 7401 et seq., in response to “dissatisfaction with the progress of existing air pollution programs.” Union Elec. Co. v. EPA, 427 U. S. 246, 249 (1976). The amendments aimed “to guarantee the prompt attainment and maintenance of specified air quality standards.” Ibid,.; D. Currie, Air Pollution §1.13, p. 1-16 (1981) (summary of 1970 amendments). Added by the 1970 amendments, §§ 108(a) and 109(a) of the Act require EPA to publish lists of emissions that “cause or contribute to air pollution which may reasonably be anticipated to endanger public health or welfare,” and to promulgate primary and secondary national ambient air quality standards (NAAQS) for such pollutants. 42 U. S. C. §§ 7408(a) and 7409(a); Whitman v. American Trucking Assns., Inc., 531 U. S. 457, 462-463 (2001). NAAQS “define [the] levels of air quality that must be achieved to protect public health and welfare.” R. Belden, Clean Air Act 6 (2001). The Agency published initial NAAQS in 1971, Union Elec., 427 U. S., at 251 (citing 40 *470CFR pt. 50 (1975)), and in 1985, NAAQS for the pollutant at issue in this case, nitrogen dioxide. 40 CFR §50.11 (2002).1

Under § 110 of the Act, also added in 1970, each State must submit for EPA approval “a plan which provides for implementation, maintenance, and enforcement of [NAAQS].” 42 U. S. C. § 7410(a)(1); cf. § 7410(c)(1) (EPA shall promulgate an implementation plan if the State’s plan is inadequate). Relevant to this case, EPA has approved Alaska’s implementation plan. 48 Fed. Reg. 30626 (1983), as amended, 56 Fed. Reg. 19288 (1991); 40 CFR § 52.96(a) (2002). To gain EPA approval, a “state implementation plan” (SIP) must “include enforceable emission limitations and other control measures, means, or techniques ... as may be necessary or appropriate to meet the applicable [CAA] requirements.” 42 U. S. C. § 7410(a)(2)(A). While States have “wide discretion” in formulating their plans, Union Elec., 427 U. S., at 250, SIPs must include certain measures Congress specified “to assure that national ambient air quality standards are achieved,” 42 U. S. C. § 7410(a)(2)(C). Among those measures are permit provisions, § 7475, basic to the administration of the program involved in this case, CAA’s “Prevention of Significant Deterioration of Air Quality” (PSD) program.

The PSD requirements, as part ments to the Act, Title I, §160 et seq., 91 Stat. 731, “are designed to ensure that the air quality in attainment areas or areas that are already 'clean’ will not degrade,” Belden, supra, at 43. See 42 U. S. C. § 7470(1) (purpose of PSD pro*471gram is to “protect public health and welfare from any. actual or potential adverse effect which in [EPA’s] judgment may reasonably be anticipate^] to occur from air pollution . . . notwithstanding attainment and maintenance of all national ambient air quality standards”). Before 1977, no CAA provision specifically addressed potential air quality deterioration in areas where pollutant levels were lower than the NAAQS. Alabama Power Co. v. Costle, 636 F. 2d 323, 346-347 (CADC 1979). Responding to litigation initiated by an environmental group,2 however, EPA issued regulations in 1974 requiring that SIPs include a PSD program. Id., at 347, and n. 18 (citing 39 Fed. Reg. 42510 (1974)). Three years later, Congress adopted the current PSD program. See S. Rep. No. 95-127, p. 11 (1977) (Congress itself has “a responsibility to delineate a policy for protecting clean air”).

The PSD program imposes on States a regime governing areas “designated pursuant to [42 U. S. C. § 7407] as attainment or unclassifiable.” § 7471.3 An attainment area is one in which the air “meets the national primary or secondary ambient air quality standard for [a regulated pollutant].” §7407(d)(l)(A)(ii). Air in an unclassifiable area “cannot be classified on the basis of available information as meeting or not meeting the national primary or secondary ambient air quality standard for the pollutant.” §7407(d)(l)(A)(iii). Northwest Alaska, the region this ease concerns, is classified as an attainment or unclassifiable area for nitrogen dioxide, 40 CFR § 81.302 (2002); therefore, the PSD program applies to emissions of that pollutant in the region. In 2002, the Agency reported that “[a]ll areas of the country that once *472violated the NAAQS for [nitrogen dioxide] now meet that standard.” EPA, Latest Findings on National Air Quality 7 (Aug. 2003).

Section 165 of the Act, 42 U. S. C. §7475, installs a permitting requirement for any “major emitting facility,” defined to include any source emitting more than 250 tons of nitrogen oxides per year, §7479(1). No such facility may be constructed or modified unless a permit prescribing emission limitations has been issued for the facility. § 7475(a)(1); see §7479(2)(C) (defining “construction” to include “modification”). Alaska’s SIP imposes an analogous requirement. 18 Alaska Admin. Code § 50.300(c)(1) (2003). Modifications to major emitting facilities that increase nitrogen oxide emissions in excess of 40 tons per year require a PSD permit. 40 CFR § 51.166(b)(23)(i) (2002); 18 Alaska Admin. Code § 50.300(h)(3)(B)(ii) (2003).

The Act sets out preconditions for the issuance of PSD permits. Inter alia, no PSD permit may issue unless “the proposed facility is subject to the best available control technology for each pollutant subject to [CAA] regulation . . . emitted from ... [the] facility.” 42 U. S. C. § 7475(a)(4). As described in the Act’s definitional provisions, “best available control technology” (BACT) means:

“an emission limitation based on the maximum degree of reduction of each pollutant subject to regulation under this chapter emitted from or which results from any major emitting facility, which the permitting authority, on a case-by-case basis, taking into account energy, environmental, and economic impacts and other costs, determines is achievable for such facility through application of production processes and available methods, systems, and techniques .... In no event shall application of ‘best available control technology’ result in emissions of any pollutants which will exceed the emissions allowed by any applicable standard established pursuant *473to section 7411 or 7412 of this title [emission standards for new and existing stationary sources].” §7479(3).

40 CFR §51.166(b)(12) (2002) (repeating statutory definition). Alaska’s SIP contains provisions that track the statutory BACT requirement and definition. 18 Alaska Admin. Code §§ 50.310(d)(3) and 50.990(13) (2003). The State, with slightly variant terminology, defines BACT as “the emission limitation that represents the maximum reduction achievable for each regulated air contaminant, taking into account energy, environmental and economic impacts, and other costs.” §50.990(13). Under the federal Act, a limited class of sources must gain advance EPA approval for the BACT prescribed in the permit. 42 U. S. C. § 7475(a)(8).

CAA also provides that a PSD permit may issue only if a source “will not cause, or contribute to, air pollution in excess of any . . . maximum allowable increase or maximum allowable concentration for any pollutant” or any NAAQS. § 7475(a)(3). Congress left to the Agency the determination of most maximum allowable increases, or “increments,” in pollutants. EPA regulations have defined increments for nitrogen oxides. 40 CFR § 51.166(c) (2002). Typically, to demonstrate that increments will not be exceeded, applicants use mathematical models of pollutant plumes, their behavior, and their dispersion. Westbrook, Air Dispersion Models: Tools to Assess Impacts from Pollution Sources, 13 Natural Resources & Env. 546, 547-548 (1999).

Among measures EPA may take to ensure compliance with the PSD program, two have special relevance here. The first prescription, § 113(a)(5) of the Act, provides that “[w]henever, on the basis of any available information, [EPA] finds that a State is not acting in compliance with any requirement or prohibition of the chapter relating to the construction of new sources or the modification of existing sources,” 42 U. S. C. § 7413(a)(5), EPA may “issue an order prohibiting the construction or modification of any major stationary source in any area to which such requirement ap*474plies,” § 7413(a)(5)(A).4 The second measure, §167 of the Act, trains on enforcement of the PSD program; it requires EPA to “take such measures, including issuance of an order, or seeking injunctive relief, as necessary to prevent the construction or modification of a major emitting facility which does not conform to the [PSD] requirements.” §7477.

B

Teck Comineo Alaska Inc. (Comineo) operates a zinc concentrate mine, the Red Dog Mine, in northwest Alaska approximately 100 miles north of the Arctic Circle and close to the native Alaskan villages of Kivalina and Noatak. App. to Pet. for Cert. 3a; Brief for Petitioner 8; Brief for Respondents 4. The mine is the region’s largest private employer. Brief for Petitioner 9. It supplies a quarter of the area’s wage base. Ibid. Comineo leases the land from the NANA Regional Corporation, an Alaskan corporation formed pursuant to the Alaska Native Claims Settlement Act, 85 Stat. 688, as amended, 43 U. S. C. § 1601 et seq. Brief for NANA Regional Corporation, Inc., as Amicus Curiae 1-2, 4.

In 1988, Comineo obtained authorization to operate the mine, a “major emitting facility” under the Act and Alaska’s SIP. App. 106. The mine’s PSD permit authorized five 5,000 kilowatt Wartsila diesel electric generators, MG-1 through MG-5, subject to operating restrictions; two of the five generators were permitted to operate only in standby status. Ibid. Petitioner Alaska Department of Environmental Conservation (ADEC) issued a second PSD permit in 1994 allowing addition of a sixth full-time generator (MG-6), removing standby status from MG-2, and imposing a new *475operational cap that allowed all but one generator to run full time. Ibid.

In 1996, Comineo initiated a project, with funding from the State, to expand zinc production by 40%. Brief for Petitioner 10; Reply Brief for Petitioner 11, n. 9. Anticipating that the project would increase nitrogen oxide emissions by more than 40 tons per year, see supra, at 472, Comineo applied to ADEC for a PSD permit to allow, inter alia, increased electricity generation by its standby generator, MG-5. App. 107-108; App. to Pet. for Cert. 33a. On March 3, 1999, ADEC preliminarily proposed as BACT for MG-5 the emission control technology known as selective catalytic reduction (SCR),5 which reduces nitrogen oxide emissions by 90%. App. 72, 108. In response, Comineo amended its application to add a seventh generator, MG-17, and to propose as BACT an alternative control technology — Low NOx6— that achieves a 30% reduction in nitrogen oxide pollutants. Brief for Respondents 5, and n. 1; App. 84.

On May 4,1999, ADEC, in conjunction with Cominco’s representative, issued a first draft PSD permit and preliminary technical analysis report that concluded Low NOx was BACT for MG-5 and MG-17. Id., at 55-95. To determine BACT, ADEC employed EPA’s recommended top-down methodology, id., at 61:

“In briefj the top-down process provides that all available control technologies be ranked in descending order of control effectiveness. The PSD applicant first examines the most stringent — or ‘top’ — alternative. That al*476ternative is established as BACT unless the applicant demonstrates, and the permitting authority in its informed judgment agrees, that technical considerations, or energy, environmental, or economic impacts justify a conclusion that the most stringent technology is not ‘achievable’ in that case. If the most stringent technology is eliminated in this fashion, then the next most stringent alternative is considered, and so on.” EPA, New Source Review Workshop Manual B.2 (Draft Oct. 1990) (hereinafter New Source Review Manual); App. 61-62.7

Applying top-down methodology, ADEC first homed in on SCR as BACT for MG-5, and the new generator, MG-17. “[W]ith an estimated reduction of 90%,” ADEC stated, SCR “is the most stringent” technology. Id., at 79. Finding SCR “technically and economically feasible,” id., at 65, ADEC characterized as “overstated” Cominco’s cost estimate of $5,643 per ton of nitrogen oxide removed by SCR, id., at 113. Using Cominco's data, ADEC reached a cost estimate running between $1,586 and $2,279 per ton. Id., at 83. Costs in that range, ADEC observed, “are well within what ADEC and EPA conside[r] economically feasible.” Id., at 84. Responding to Cominco’s comments on the preliminary permit, engineering staff in ADEC’s Air Permits Program pointed out that, according to information Comineo provided to ADEC, “SCR has been installed on similar diesel-fired engines throughout the world.” Id., at 102.

Despite its staff’s clear view “that SCR (the most effective individual technology) [was] technologically, environmentally, and economically feasible for the Red Dog power plant engines,” id., at 103-104, ADEC endorsed the alternative prof*477fered by Comineo. To achieve nitrogen oxide emission reductions commensurate with SCR’s 90% impact, Comineo proposed fitting the new generator MG-17 and the six existing generators with Low NOx. Ibid.8, Comineo asserted that it could lower net emissions by 396 tons per year if it fitted all seven generators with Low NOx rather than fitting two (MG-5 and MG-17) with SCR and choosing one of them as the standby unit. Id., at 87. Cominco’s proposal hinged on the “assumption . . . that under typical operating conditions one or more engines will not be running due to maintenance of standby-generation capacity.” Ibid. If all seven generators ran continuously, however, Cominco’s alternative would increase emissions by 79 tons per year. Ibid. Accepting Cominco’s submission, ADEC stated that Cominco’s Low NOx solution “achieve[d] a similar maximum NOx reduction as the most stringent controls; [could] potentially result in a greater NOx reduction; and is logistically and economically less onerous to Comineo.” Id., at 87-88.

On the final day of the public comment period, June 2, 1999, the United States Department of the Interior, National Parks Service (NPS), submitted comments to ADEC. App. to Pet. for Cert. 33a; App. 97, 108. NPS objected to the projected offset of new emissions from MG-5 and MG-17 against emissions from other existing generators that were not subject to BACT. Letter from John Notar, NPS Air Resources Division, to Jim Baumgartner, ADEC (June 2, 1999). Such an offset, NPS commented, “is neither allowed by BACT, nor achieves the degree of reduction that would result if all the generators that are subject to BACT were equipped with SCR.” Id., at 3. NPS further observed that the proposed production-increase project would remove operating restrictions that the 1994 PSD permit had placed on four of the existing generators — MG-1, MG-3, MG-4, *478and MG-5. App. to Pet. for Cert. 34a. Due to that alteration, NPS urged, those generators, too, became part of the production-expansion project and would be subject to the BACT requirement. Ibid.

Following NPS’ lead, EPA wrote to ADEC on July 29, 1999, commenting: “Although ADEC states in its analysis that [SCR], the most stringent level of control, is economically and technologically feasible, ADEC did not propose to require SCR. . . . [0]nce it is determined that an emission unit is subject to BACT, the PSD program does not allow the imposition of a limit that is less stringent than BACT.” App. 96-97. A permitting authority, EPA agreed with NPS, could not offset new emissions “by imposing new controls on other emission units” that were not subject to BACT, Id., at 97. New emissions could be offset only against reduced emissions from sources covered by the same BACT authorization. Id., at 285-286. EPA further agreed with NPS that, based on the existing information, BACT would be required for MG-1, MG-3, MG-4, and MG-5. Id., at 97.

After receiving EPA comments, ADEC issued a second draft PSD permit and technical analysis report on September 1,1999, again finding Low NOx to be BACT for MG-17. Id., at 105-117. Abandoning the emissions-offsetting justification advanced in the May 4 draft permit, ADEC agreed with NPS and EPA that “emission reductions from sources that were not part of the permit action,” here MG-1, MG-2, MG-3, MG-4, MG-5, and MG-6, could not be considered in determining BACT for MG-17. Id., at 111; id., at 199 (same).9

*479ADEC conceded that, lacking data from Comineo, it had made “no judgment... as to the impact of ... [SCR] on the operation, profitability, and competitiveness of the Red Dog Mine.” Id., at 116. Contradicting its May 1999 conclusion that SCR was “technically and economically feasible,” see supra, at 476, ADEC found in September 1999 that SCR imposed “a disproportionate cost” on the mine. App. 116. ADEC concluded, on a “cursory review,” that requiring SCR for a rural Alaska utility would lead to a 20% price increase, and that in comparison with other BACT technologies, SCR came at a “significantly higher” cost. Ibid. No economic basis for a comparison between the mine and a rural utility appeared in ADEC’s technical analysis.

EPA protested the revised permit. In a September 15, 1999, letter, the Agency stated: “Comineo has not adequately demonstrated any site-specific factors to support their claim that the installation of [SCR] is economically infeasible at the Red Dog Mine. Therefore, elimination of SCR as BACT based on cost-effectiveness grounds is not supported by the record and is clearly erroneous.” Id., at 127; see id., at 138 (ADEC’s record does not support the departure from ADEC’s initial view that the costs for SCR were economically feasible).

To justify the September 1, 1999, permit, EPA suggested, ADEC could “include an analysis of whether requiring Com-ineo to install and operate [SCR] would have any adverse economic impacts upon Comineo specifically.” Id., at 127. Stating that such an inquiry was unnecessary and expressing “concerns related to confidentiality,” Comineo declined to submit financial data. Id., at 134. In this regard, Comineo simply asserted, without detail, that the company’s “overall debt remains quite high” despite continuing profits. Id., at *480134-135. Comineo also invoked the need for “[¡Industrial development in rural Alaska.” Id., at 135.

On December 10, 1999, ADEC issued the final permit and technical analysis report. Once again, ADEC approved Low NOx as BACT for MG-17 “[t]o support Cominco’s Red Dog Mine Production Rate Increase Project, and its contributions to the region.” Id., at 208. ADEC did not include the economic analysis EPA had suggested. Id., at 152-246. Indeed, ADEC conceded again that it had made “no judgment ... as to the impact of... [SCR’s] cost on the operation, profitability, and competitiveness of the Red Dog Mine.” Id., at 207. Nonetheless, ADEC advanced, as cause for its decision, SCR’s adverse, effect on the mine’s “unique and continuing impact on the economic diversity of th[e] region” and on the venture’s “world competitiveness.” Id., at 208. ADEC did not explain how its inferences of adverse effects on the region’s economy or the mine’s “world competitiveness” could be made without financial information showing SCR’s impact on the “operation, profitability, and competitiveness” of the mine. Id., at 207, 299. Instead, ADEC reiterated its rural Alaska utility analogy, and again compared SCR’s cost to the costs of other, less stringent, control technologies. Id., at 205-207.

The same day, December 10, 1999, EPA issued an order to ADEC, under §§ 113(a)(5) and 167 of the Act, 42 U. S. C. §§ 7413(a)(5) and 7477, prohibiting ADEC from issuing a PSD permit to Comineo “unless ADEC satisfactorily documents why SCR is not BACT for the Wartsila diesel generator [MG-17].” App. to Pet. for Cert. 36a. In the letter accompanying the order, the Agency stated that “ADEC’s own analysis supports the determination that BACT is [SCR], and that ADEC’s decision in the proposed permit therefore is both arbitrary and erroneous.” App. 149.

On February 8, 2000, EPA, again invoking its authority under §§ 113(a)(5) and 167 of the Act, issued a second order, *481this time prohibiting Comineo from beginning “construction or modification activities at the Red Dog mine.” App. to Pet. for Cert. 49a. A third order, issued on March 7, 2000, superseding and vacating the February 8 order, generally prohibited Comineo from acting on ADEC’s December 10 PSD permit but allowed limited summer construction. Id., at 62a-64a. On April 25, 2000, EPA withdrew its December 10 order. App. 300; App. to Pet. for Cert. 6a. Once ADEC issued the permit, EPA explained, that order lacked utility. On July 16, 2003, ADEC granted Comineo a PSD permit to construct MG-17 with SCR as BACT. Letter from Theodore B. Olson, Solicitor General, to William K. Suter, Clerk of the Court (Aug. 21, 2003). Under the July 16, 2003, permit, SCR ceases to be BACT “if and when the case currently pending before the Supreme Court of the United States of America is decided in favor of the State of Alaska.” ADEC, Air Quality Control Construction Permit, Final Technical Analysis Report, Permit No. 9932-AC005, Revision 2, p. 7.

The day EPA issued its first order against Comineo, February 8, 2000, ADEC and Comineo petitioned the Court of Appeals for the Ninth Circuit for review of EPA’s orders. App. 11. The Agency initially moved to dismiss, urging that the Court of Appeals lacked subject-matter jurisdiction. In an order released March 27, 2001, the Ninth Circuit concluded that it had adjudicatory authority pursuant to 42 U. S. C. § 7607(b)(1), which lodges jurisdiction over challenges to “any . . . final [EPA] action” in the Courts of Appeals. Alaska v. United States EPA, 244 F. 3d 748, 750-751.10

The Court of Appeals resolved the merits in a judgment released July 30, 2002. 298 F. 3d 814 (CA9). It held that *482EPA had authority under §§ 113(a)(5) and 167 to issue the contested orders, and that the Agency had properly exercised its discretion in doing so. Id., at 820-823. Concerning EPA’s authority under §§ 113(a)(5) and 167, the Court of Appeals observed first that “the question presented is what requirements the state must meet” under the Act to issue a PSD permit, not what the correct BACT might be. Id., at 821 (emphasis in original). Concluding that EPA had “authority to determine the reasonableness or adequacy of the state's justification for its decision,” the Court of Appeals emphasized that the “provision of a reasoned justification” by a permitting authority is undeniably a “requirement” of the Act. Ibid. EPA had properly exercised its discretion in issuing the three orders, the Ninth Circuit ultimately determined, because (1) Comineo failed to “demonstrate] that SCR was economically infeasible,” and (2) “ADEC failed to provide a reasoned justification for its elimination of SCR as a control option.” Id., at 823. We granted certiorari, 537 U. S. 1186 (2003), to resolve an important question of federal . law, i. e., the scope of EPA’s authority under §§ 113(a)(5) and 167, and now affirm the Ninth Circuit's judgment.

II

ADEC contested EPA’s orders under 42 U. S. C. § 7607 (b)(1), which renders reviewable in the appropriate federal court of appeals any EPA “final action." Before the Ninth Circuit, EPA unsuccessfully urged that its orders were “interlocutory,” and therefore unreviewable in court unless and until EPA chose to commence an enforcement action.11 A preenforcement contest could be maintained in the Court of Appeals under § 7607(b)(1), the Ninth Circuit held, for in the circumstances presented, EPA’s actions had the requisite finality.

*483It was undisputed, the Court of Appeals observed, that EPA had spoken its “‘last word”’ on whether ADEC had adequately justified its conclusion that Low NOx was the best available control technology for the MG-17 generator. 244 F. 3d, at 750. Further, EPA’s orders effectively halted construction of the MG-17 generator, for Comineo would risk civil and criminal penalties if it defied a valid EPA directive.

In this Court, EPA agrees with the Ninth Circuit’s finality determination. See Brief for Respondents 16-20; Tr. of Oral Arg. 43-44. We are satisfied that the Court of Appeals correctly applied the guides we set out in Bennett v. Spear, 520 U. S. 154, 177-178 (1997) (to be “final,” agency action must “mark the ‘consummation’ of the agency’s decisionmaking process,” and must either determine “rights or obligations” or occasion “legal consequences” (some internal quotation marks omitted)). As the Court of Appeals stated, EPA had “asserted its final position on the factual circumstances” underpinning the Agency’s orders, 244 F. 3d, at 750, and if EPA’s orders survived judicial review, Cominco could not escape the practical and legal consequences (lost costs and vulnerability to penalties) of any ADEC-permitted construction Cominco endeavored, ibid.

No question has been raised here, we note, about the adequacy of EPA’s preorder procedures under the Due Process Clause or the Administrative Procedure Act. Cf. Vermont Yankee Nuclear Power Corp. v. Natural Resources Defense Council, Inc., 435 U. S. 519, 544 (1978) (agencies have authority to “fashion their own rules of procedure,” even when a statute does not specify what process to use). Furthermore, in response to ADEC’s initial contention that the record was incomplete, the Ninth Circuit gave EPA an opportunity to supplement the record, and thereafter obtained from all parties agreement “that the record as it stood was adequate to resolve [ADEC’s review petition].” 298 F. 3d, at 818.

*484III

A

Centrally at issue in this case is the question whether EPA’s oversight role, described by Congress in CAA §§ 113(a)(5) and 167, see supra, at 473-474, extends to ensuring that a state permitting authority’s BACT determination is reasonable in light of the statutory guides. Sections 113(a)(5) and 167 lodge in the'Agency encompassing supervisory responsibility over the construction and modification of pollutant emitting facilities in areas covered by the PSD program. 42 U. S. C. §§ 7413(a)(5) and 7477. In notably capacious terms, Congress armed EPA with authority to issue orders stopping construction when “a State is not acting in compliance with any [CAA] requirement or prohibition . . . relating to the construction of new sources or the modification of existing sources,” § 7413(a)(5), or when “construction or modification of a major emitting facility . . . does not conform to the requirements of [the PSD program],” § 7477.

The federal Act enumerates several “^reconstruction requirements” for the PSD program. §7475. Absent these, “[n]o major emitting facility .. . may be constructed.” Ibid. One express preconstruction requirement is inclusion of a BACT determination in a facility’s PSD permit. §§ 7475(a) (1) and (4). As earlier set out, see supra, at 472, the Act defines BACT as “an emission limitation based on the maximum degree of reduction of [a] pollutant . . . which the permitting authority, on a case-by-case basis, taking into account energy, environmental, and economic impacts and other costs, determines is achievable for [a] facility.” §7479(3). Under this formulation, the permitting authority, ADEC here, exercises primary or initial responsibility for identifying BACT in line with the Act’s definition of that term.

All parties agree that one of the “many requirements in the PSD provisions that the EPA may enforce” is “that a *485[PSD] permit contain a BACT limitation.” Brief for Petitioner 34; see id., at 22, 25 (same). See also Brief for Respondents 23. It is therefore undisputed that the Agency may issue an order to stop a facility’s construction if a PSD permit contains no BACT designation.

EPA reads the Act’s definition of BACT, together with CAA’s explicit listing of BACT as a “^reconstruction re-quiremen[t],” to mandate not simply a BACT designation, but a determination of BACT faithful to the statute’s definition. In keeping with the broad oversight role §§ 113(a)(5) and 167 vest in EPA, the Agency maintains, it may review permits to ensure that a State’s BACT determination is reasonably moored to the Act’s provisions. See id., at 24. We hold, as elaborated below, that the Agency has rationally construed the Act’s text and that EPA’s construction warrants our respect and approbation.

BACT’s statutory definition requires selection of an emission control technology that results in the “maximum” reduction of a pollutant “achievable for [a] facility” in view of “energy, environmental, and economic impacts and other costs.” 42 U. S. C. § 7479(3). This instruction, EPA submits, cabins state permitting authorities’ discretion by granting only “authority to make reasonable BACT determinations,” Brief for Respondents 27 (emphasis in original), i. e., decisions made with fidelity to the Act’s purpose “to insure that economic growth will occur in a manner consistent with the preservation of existing clean air resources,” 42 U. S. C. § 7470(3). Noting that state permitting authorities’ statutory discretion is constrained by CÁA’s strong, normative terms “maximum” and “achievable,” §7479(3),12 EPA reads §§ 113(a)(5) *486and 167 to empower the federal Agency to check a state agency's unreasonably lax BACT designation. See Brief for Respondents 27.

EPA stresses Congress' reason for enacting the PSD program — to prevent significant deterioration of air quality in clean-air areas within a State and in neighboring States. §§ 7470(3), (4); see id., at 33. That aim, EPA urges, is unlikely to be realized absent an EPA surveillance role that extends to BACT determinations. The Agency notes in this regard a House Report observation:

“Without national guidelines for the prevention of significant deterioration a State deciding to protect its clean air resources will face a double threat. The prospect is very real that such a State would lose existing industrial plants to more permissive States. But additionally the State will likely become the target of 'economic-environmental blackmail’ from new industrial plants that will play one State off against another with threats to locate in whichever State adopts the most permissive pollution controls.” H. R. Rep. No. 95-294, p. 134 (1977).

The House Report further observed that “a community that sets and enforces strict standards may still find its air polluted from sources in another community or another State.” Id., at 135 (quoting 116 Cong. Rec. 32909 (1970)). Federal Agency surveillance of a State’s BACT designation is needed, EPA asserts, to restrain the interjurisdictional pressures to which Congress was alert. See Brief for Respondents 33-34, 43; Brief for Vermont et al. as Amici Curiae 12 (“If EPA has authority to ensure a reasonable level of consistency among BACT determinations nationwide, then every State can feel more confident about maintaining stringent standards without fear of losing its current industry or alienating prospective industry.”).

*487The CAA construction EPA advances in this litigation is reflected in interpretive guides the Agency has several times published. See App. 268-269 (1983 EPA PSD guidance memorandum noting the Agency’s “oversight function”); id., at 274 (1988 EPA guidance memorandum stating EPA may find a BACT determination deficient if it is “not based on a reasoned analysis”); id., at 281-282 (1993 guidance memorandum stating that “EPA acts to ensure that the state exercises its discretion within the bounds of the law” (internal quotation marks omitted); as to BACT, EPA will not intervene if the state agency has given “a reasoned justification for the basis of its decision” (internal quotation marks omitted)). See also Approval and Promulgation of Air Quality Implementation Plans; Commonwealth of Virginia- — Prevention of Significant Deterioration Program, 63 Fed. Reg. 13797 (1998) (EPA will “review whether any determination by the permitting authority was made on reasonable grounds properly supported on the record, described in enforceable terms, and consistent with all applicable requirements”). We “normally accord particular deference to an agency interpretation of ‘longstanding’ duration,” Barnhart v. Walton, 535 U. S. 212, 220 (2002) (quoting North Haven Bd. of Ed. v. Bell, 456 U. S. 512, 522, n. 12 (1982)), recognizing that “well-reasoned views” of an expert administrator rest on “ ‘a body of experience and informed judgment to which courts and litigants may properly resort for guidance,’ ” Bragdon v. Abbott, 524 U. S. 624, 642 (1998) (quoting Skidmore v. Swift & Co., 323 U. S. 134, 139-140 (1944)).

We have previously accorded dispositive effect to EPA’s interpretation of an ambiguous CAA provision. See Chevron U. S. A. Inc. v. Natural Resources Defense Council, Inc., 467 U. S. 837, 865-866 (1984); Union Elec., 427 U. S., at 256. The Agency’s interpretation in this case, presented in internal guidance memoranda, however, does not qualify for the dispositive force described in Chevron. See Christensen v. Harris County, 529 U. S. 576, 587 (2000) (“Interpretations *488such as those in .. . policy statements, agency manuals, and enforcement guidelines, all of which lack the force of law— do not warrant Chevron-style deference.”); accord United States v. Mead Corp., 533 U. S. 218, 234 (2001). Cogent “administrative interpretations .. . not [the] products of formal rulemaking ... nevertheless warrant respect.” Washington State Dept, of Social and Health Servs. v. Guardianship Estate of Keffeler, 537 U. S. 371, 385 (2003). We accord EPA’s reading of the relevant statutory provisions, §§ 7413(a)(5), 7470(3), 7470(4), 7475(a)(4), 7477, and 7479(3), that measure of respect.

B

' ADEC assails the Agency’s construction of the Act on several grounds. Its arguments do not persuade us to reject as impermissible EPA’s longstanding, consistently maintained interpretation.

ADEC argues that the statutory § 7479(3), unambiguously assigns to “the permitting authority” alone determination of the control technology qualifying as “best available.” Brief for Petitioner 21-26. Because the Act places responsibility for determining BACT with “the permitting authority,” ADEC urges, CAA excludes federal Agency surveillance reaching the substance of the BACT decision. Id., at 22-25. EPA’s enforcement role, ADEC maintains, is restricted to the requirement “that the permit contain a BACT limitation.” Id., at 34.

Understandably, Congress authorities with initial responsibility to make BACT determinations “case-by-case.” §7479(3). A state agency, no doubt, is best positioned to adjust for local differences in raw materials or plant configurations, differences that might make a technology “unavailable” in a particular area. But the fact that the relevant statutory guides — “maximum” pollution reduction, considerations of energy, environmental, and economic impacts — may not yield a “single, objectively ‘correct’ BACT determination,” id., at 23, surely does not *489signify that there can be no unreasonable determinations. Nor does Congress’ sensitivity to site-specific factors necessarily imply a design to preclude in this context meaningful EPA oversight under §§ 113(a)(5) and 167. EPA claims no prerogative to designate the correct BACT; the Agency asserts only the authority to guard against unreasonable designations. See 298 F. 3d, at 821 (“the question presented is what requirements the state must meet,” not what final substantive decision the State must make (emphasis in original)).13

Under ADEC’s interpretation, EPA properly inquires whether a BACT determination appears in a PSD permit, *490Brief for Petitioner 34, but not whether that BACT determination “was made on reasonable grounds properly supported on the record,” 63 Fed. Reg., at 13797. Congress, however, vested EPA with explicit and sweeping authority to enforce CAA “requirements” relating to the construction and modification of sources under the PSD program, including BACT. We fail to see why Congress, having expressly endorsed an expansive surveillance role for EPA in two independent CAA provisions, would then implicitly preclude the Agency from verifying substantive compliance with the BACT provisions and, instead, limit EPA’s superintendence to the insubstantial question whether the state permitting authority had uttered the key words “BACT.”

We emphasize, however, that EPA’s rendition of the Act’s less than crystalline text leaves the “permitting authority” considerable leeway. The Agency acknowledges “the need to accord appropriate deference” to States’ BACT designations,. Brief for Respondents 43, and disclaims any intention to “‘second guess’ state decisions,” 63 Fed. Reg., at 13797. Only when a state agency’s BACT determination is “not based on a reasoned analysis,” App. 274, may EPA step in to ensure that the statutory requirements are honored.14 EPA *491adhered to that limited role here, explaining why ADEC’s BACT determination was “arbitrary” and contrary to ADEC’s own findings. Id., at 149-150. EPA’s limited but vital role in enforcing BACT is consistent with a scheme that “places primary responsibilities and authority with the States, backed by the Federal Government.” S. Rep. No. 95-127, p. 29.

ADEC also points to 42 U. S. C. § 7475(a)(8), a provision of the Act expressly requiring, in a limited category of cases, EPA approval of a state permitting authority’s BACT determination before a facility may be constructed. See Brief for Petitioner 25; Reply Brief for Petitioner 6. Had Congress intended EPA superintendence of BACT determinations, ADEC urges, Congress would have said so expressly by mandating Agency approval of all, not merely some, BACT determinations. Brief for Petitioner 25-26. ADEC’s argument overlooks the obvious difference between a statutory requirement, e. g., § 7475(a)(8), and a statutory authorization. Sections 113(a)(5) and 167 sensibly do not require EPA approval of all state BACT determinations, they simply authorize EPA to act in the unusual case in which a state permitting authority has determined BACT arbitrarily. EPA recognizes that its authorization to issue a stop order may be exercised only when a state permitting authority’s decision is unreasonable; in contrast, a required approval may be withheld if EPA would come to a different determination on the merits. See, e. g., 57 Fed. Reg. 28095 (1992) (“EPA acknowledges that states have the primary role in administering and enforcing the various components of the PSD program. States have been largely successful in this effort, and EPA’s involvement in interpretative and enforcement issues is limited to only a small number of cases.”).

Even if the Act imposes a requirement of reasoned justification for a BACT determination, ADEC ultimately argues, such a requirement may be enforced only through state ad*492ministrative and judicial processes. Brief for Petitioner 34-38.15 State review of BACT decisions, according to ADEC, allows development of an adequate factual record, properly imposes the burden of persuasion on EPA when it challenges a State’s BACT determination, and promotes certainty. Id., at 36-37. Unless EPA review of BACT determinations is channeled into state administrative and judicial forums, ADEC suggests, “there is nothing to prevent the EPA from invalidating a BACT determination at any time — months, even years, after a permit has been issued.” Id., at 35.

It would be unusual, to say the least, for Congress to remit a federal agency enforcing federal law solely to state court. We decline to read such an uncommon regime into the Act’s silence. EPA, the expert federal agency charged with enforcing the Act, has interpreted the BACT provisions and its own §§ 113(a)(5) and 167 enforcement powers not to require recourse to state processes before stopping a facility’s con*493struction. See supra, at 485-488. That rational interpretation, we agree, is surely permissible.16

Nor are we persuaded by ADEC’s practical concerns. We see no reason to conclude that an appropriate record generally cannot be developed to allow informed federal-court review when EPA disputes a BACT decision’s reasonableness. ADEC contends that, in this very case, “the State’s BACT determination was reviewed by the Ninth Circuit on an incomplete record.” Brief for Petitioner 37. ADEC, however, offers no particulars to back up its assertion that the Court of Appeals proceeded on. an inadequate evidentiary record. We note again that the Ninth Circuit ordered EPA to submit a complete administrative record. 298 F. 3d, at 818. After the Agency declared that the record was complete, “all the parties effectively agreed that the record as it stood was adequate to resolve the issues on appeal.” Ibid.

As to the burdens of production and persuasion, nothing in the Act suggests that EPA gains a proof-related tactical advantage by issuing a stop-construction order instead of seeking relief through a civil action. But cf. post, at 510 (EPA authority to issue stop-construction orders creates “the anomaly of shifting the burden of pleading and of initiating litigation from EPA to the State”). Correspondingly, nothing in our decision today invites or permits EPA to achieve an unfair advantage through its choice of litigation forum. In granting EPA a choice between initiating a civil action and exercising its stop-construction-order authority, see supra, at 473-474, 492, n. 15, Congress nowhere suggested that the allocation of proof burdens would differ depending upon which enforcement route EPA selected. The *494point ought not to be left in doubt. Accordingly, we hold that in either an EPA-initiated civil action or a challenge to an EPA stop-construction order filed in state or federal court, the production and persuasion burdens remain with EPA and the underlying question a reviewing court resolves remains the same: Whether the state agency’s BACT determination was reasonable, in light of the statutory guides and the state administrative record. See supra, at 485-486, 491.17

The Ninth Circuit’s review of EPA’s order is in keeping with our holding that EPA may not reduce the burden it must carry by electing to invoke its 'stop-construction-order authority. Specifically, the Court of Appeals rested its judgment on what EPA showed from ADEC’s own report: “(1) Comineo failed to meet its burden of demonstrating [to ADEC] that SCR was economically infeasible; and (2) ADEC failed to provide a reasoned justification for its elimination of SCR as a control option.” 298 F. 3d, at 823. EPA’s conclusions, and the basis for them, support the Court of Appeals’ determination that the federal Agency’s grounds for issuing the orders under review were not “arbitrary] and capriciou[s].” Ibid. Our own analysis, infra, at 497-502, similarly hinges on the question whether ADEC’s BACT determination was a reasonable one. Our analysis would have *495taken the same path had EPA initiated a civil action pursuant to § 113(a)(5)(C), or if the suit under consideration had been filed initially in state court.

Nor do we find compelling ADEC’s suggestion, reiterated by the dissent, that, if state courts are not the exclusive judicial arbiters, EPA would be free to invalidate a BACT determination “months, even years, after a permit has been issued.” Brief for Petitioner 35; post, at 512-514. This case threatens no such development. It involves preconstruction-orders issued by EPA, see supra, at 481, not postconstruction federal Agency directives. EPA itself regards it as “imperative” to act on a timely basis, recognizing that courts are “less likely to require new sources to accept more stringent permit conditions the farther planning and construction have progressed.” App. 273 (July 15, 1988, EPA guidance memorandum). In the one instance of untimely EPA action ADEC identifies, the federal courts declined to permit enforcement to proceed. See United States v. AM General Corp., 34 F. 3d 472, 475.(CA7 1994) (affirming District Court’s dismissal of an EPA-initiated enforcement action where EPA did not act until well after the facility received a PSD permit and completed plant modifications). EPA, we are confident, could not indulge in the inequitable conduct ADEC and the dissent hypothesize while the federal courts sit to review EPA’s actions, Cf. Walz v. Tax Comm’n of City of New York, 397 U. S. 664, 678-679 (1970); Panhandle Oil Co. v. Mississippi ex rel. Knox, 277 U. S. 218, 223 (1928) (Holmes, J., dissenting), overruled in part by Alabama v. King & Boozer, 314 U. S. 1, 8-9 (1941).

In sum, EPA interprets the Act to allow substantive federal Agency surveillance of state permitting authorities’ BACT determinations subject to federal-court review. We credit EPA’s longstanding construction of the Act and confirm EPA’s authority, pursuant to §§ 113(a)(5) and 167, to rule on the reasonableness of BACT decisions by state permitting authorities.

*496IV

A

We turn finally, and more particularly, to the reasons why we conclude that EPA properly exercised its statutory authority in this case. ADEC urges that, even if the Act allows the Agency to issue stop-construction orders when a state permitting authority unreasonably determines BACT, EPA acted impermissibly in this instance. See Brief for Petitioner 39-48. We note, first, EPA’s threshold objection. ADEC’s petition to this Court questioned whether the Act accorded EPA oversight authority with respect to a State’s BACT determination. Pet. for Cert. 13-22. ADEC did not present, as a discrete issue, the question whether EPA, assuming it had authority to review the substance of a state BACT determination, nevertheless abused its authority by countermanding ADEC’s permit for the Red Dog Mine expansion. See Brief for Respondents 44-45; cf. Reply Brief for Petitioner 15-16, n. 12 (“EPA asserts authority to overturn only ‘arbitrary or unreasoned’ state BACT determinations. . . . Thus, whether the State issued a reasoned justification is ‘fairly included’ within the question presented[J”). Treating the case-specific issue as embraced within the sole question presented, we are satisfied that EPA did not act arbitrarily in finding that ADEC furnished no tenable accounting for its determination that Low NOx was BACT for MG-17.

Because the Act itself does not specify a standard for judicial review in this instance,18 we apply the familiar default standard of the Administrative Procedure Act, 5 U. S. C. § 706(2)(A), and ask whether the Agency’s action was “arbi*497trary, capricious, an abuse of discretion, or otherwise not in accordance with law.” Even when an agency explains its decision with “less than ideal clarity,” a reviewing court will not upset the decision on that account “if the agency’s path may reasonably be discerned.” Bowman Transp., Inc. v. Arkansas-Best Freight System, Inc., 419 U. S. 281, 286 (1974). EPA’s three skeletal orders to ADEC and Comineo surely are not composed with ideal clarity. These orders, however, are properly read together with accompanying explanatory correspondence from EPA; so read, the Agency’s comments and orders adequately ground the determination that ADEC’s acceptance of Low NOx for MG-17 was unreasonable given the facts ADEC found.

In the two draft permits and the final permit, ADEC formally followed the EPA-recommended top-down methodology to determine BACT, as Cominco had done in its application. App. 61, 109, 175; see supra, at 475-476. Employing that methodology in the May 1999 draft permit, ADEC first concluded that SCR was the most stringent emission-control technology that was both “technically and economically feasible.” App. 65; see supra, at 476. That technology should have been designated BACT absent “technical considerations, or energy, environmental, or economic impacts justifying] a conclusion that [SCR was] not ‘achievable’ in [this] case.” New Source Review Manual, p. B.2; App. 61-62. ADEC nevertheless selected Low NOx as BACT; ADEC did so in May 1999 based on Cominco’s suggestion that fitting all Red Dog Mine generators with Low NOx would reduce aggregate emissions. Id., at 87, 111-112; see supra, at 476-477.

In September and December 1999, ADEC again rejected SCR as BACT but no longer relied on Cominco’s suggestion that it could reduce aggregate emissions by equipping all generators with Low NOx. See supra, at 478-480. ADEC candidly stated that it aimed “[t]o support Cominco’s Red Dog Mine Production Rate Increase Project, and its contri*498butions to the region.” App. 208. In these second and third rounds, ADEC rested its selection of Low NOx squarely and solely on SCR’s “disproportionate cost.” Id., at 116; id., at 112-117, 203-208; supra, at 478-480.

EPA concluded that ADEC’s switch from finding SCR economically feasible in May 1999 to finding SCR economically infeasible in September 1999 had no factual basis in the record. See App: 138. In the September and December 1999 technical analyses, ADEC acknowledged that “no judgment [could then] be made as to the impact of [SCR’s] cost on the operation, profitability, and competitiveness of the Red Dog Mine.” Id., at 116, 207. ADEC nevertheless concluded that SCR would threaten both the Red Dog Mine’s “unique and continuing impact on the economic diversity” of northwest Alaska and the mine’s “world competitiveness.” Id., at 208. ADEC also stressed the mine’s role as employer in an area with “historical high unemployment and limited permanent year-round job opportunities.” Id., at 207.

We do not see how ADEC, having acknowledged that no determination “[could], be made as to the impact of [SCR’s] cost on the operation . . . and competitiveness of the [mine],” ibid., could simultaneously proffer threats to the mine’s operation or competitiveness as reasons for declaring SCR economically infeasible. ADEC, indeed, forthrightly explained why it was disarmed from reaching any judgment on whether, or to what extent, implementation of SCR would adversely affect the mine’s operation or profitability: Com-ineo had declined to provide the relevant financial data, disputing the need for such information and citing “confidentiality” concerns, id., at 134; see supra, at 479-480; 298 F. 3d, at 823 (“Comineo failed to meet its burden of demonstrating that SCR was economically infeasible.”). No record evidence suggests that the mine, were it to use SCR for its new generator, would be obliged to cut personnel or raise zinc prices. Absent evidence of that order, ADEC lacked cause *499for selecting Low NOx as BACT based on the more stringent control’s impact on the mine’s operation or competitiveness.

Nor has ADEC otherwise justified its choice of Low NOx. To bolster its assertion that SCR was too expensive, ADEC invoked four BACT determinations made in regard to diesel generators used for primary power production; BACT’s cost, in those instances, ranged from $0 to $936 per ton of nitrogen oxide removed. App. 205-206; supra, at 480. ADEC itself, however, had previously found SCR’s per-ton cost, then estimated as $2,279, to be “well within what ADEC and EPA considers economically feasible.” App. 84; cf. id., at 204 (estimating SCR’s per ton cost to be $2,100). No reasoned explanation for ADEC’s retreat from this position appears in the final permit. See id., at 138 (“[SCR’s cost falls] well within the range of costs EPA has seen permitting authorities nationwide accept as economically feasible for NOx control except where there are compelling site specific factors that indicate otherwise.”). Tellingly, as to examples of low-cost BACT urged by Comineo, ADEC acknowledged: “The cited examples of engines permitted in Alaska without requiring SCR are not valid examples as they either took place over 18 months ago or were not used for similar purposes.” Id., at 233-234 (footnote omitted). ADEC added that it has indeed “permitted [Alaska] projects requiring SCR.” Id., at 234. Further, EPA rejected ADEC’s comparison between the miné and a nina] utility, séé supra, at 479, because "no facts exist to suggest that the ‘economic impact’ of the incrementally higher cost of SCR on the world’s largest producer of zinc concentrates would be anything like its impact on a rural, non-profit utility that must pass costs on to a small base of individual consumers,” Brief for Respondents 49; App. 138-139 (similar observation in Nov. 10, 1999, EPA letter).

ADEC’s basis for selecting Low NOx thus reduces to a readiness “[t]o support Cominco’s Red Dog Mine Production Rate Increase Project, and its contributions to the region.” *500Id., at 208. This justification, however, hardly meets ADEC’s own standard of a "source-specific ... economic impact] which demonstrate^] [SCR] to be inappropriate as BACT.” Id., at 177. In short, as the Ninth Circuit determined, EPA validly issued stop orders because ADEC’s BACT designation simply did not qualify as reasonable in light of the statutory guides.

In its briefs to this Court, ADEC nonetheless justifies its selection of Low NOx as BACT for MG-17 on the ground that lower aggregate emissions would result from Cominco’s “agree[ment] to install Low NOx on all its generators.” Brief for Petitioner 42, and n. 12 (emphasis added); id., at 29; Reply Brief for Petitioner 19, n. 16. We need not dwell on ADEC’s attempt to resurrect Cominco’s emissions-offsetting suggestion, see supra, at 477, adopted in the initial May 1999 draft permit, but thereafter dropped. As ADEC acknowledges, the final PSD permit did not offset MG-17’s emissions against those of the mine’s six existing generators, installations that were not subject to BACT. Brief for Petitioner 42, n. 12; App. 149. ADEC recognized in September and December 1999 that a State may treat emissions from several pollutant sources as falling under one “bubble”19 for PSD permit purposes only if every pollutant source so aggregated is “part of the permit action.” Id., at 111, 199. Offsetting new emissions against those from any of the mine’s other generators, ADEC agreed, “[was] not a consideration of the BACT review provided for by the applicable law or guidelines,” for those generators remained outside the permit’s compass. Id., at 112,199. ADEC plainly did not, and could not, base its December 10, 1999, permit and technical analysis on an emissions-offsetting rationale drawing in gen*501erators not subject to BACT. Id., at 111-112.20 By that time, only MG-17 was “part of the permit action.” Id., at 111, 199.

B

We emphasize that today’s disposition does not impede ADEC from revisiting the BACT determination in question. In letters and orders throughout the permitting process, EPA repeatedly commented that it was open to ADEC to prepare “an appropriate record” supporting its selection of Low NOx as BACT. Tr. of Oral Arg. 35; see App. 127 (attachment to Sept. 28,1999, EPA letter to ADEC, stating “an analysis of whether requiring Comineo to install and operate [SCR] would have any adverse economic impacts upon Com-ineo specifically” might demonstrate SCR’s economic infeasibility); id., at 150 (letter accompanying EPA’s Dec. 10, 1999, finding of noncompliance and order reiterating the Agency’s willingness to “review, and consider any additional information or analyses provided by ADEC or Comineo” on Low NOx as BACT); App. to Pet. for Cert. 36a (EPA Dec. 10, 1999, order inviting ADEC to justify its choice of Low NOx by “documenting] why SCR is not BACT [for MG-17]”); id., at 49a (similar statement in Feb. 8, 2000, order). At oral argument, counsel for EPA reaffirmed that, “absolutely,” ADEC could reconsider the matter and, on an “appropriate *502record,” endeavor to support Low NOx as BACT. Tr. of Oral Arg. 35.21 We see no reason not to take EPA at its word.

* * *

In sum, we conclude that EPA has supervisory authority over the reasonableness of state permitting authorities’ BACT determinations and may issue a stop-construction order, under §§ 113(a)(5) and 167, if a BACT selection is not reasonable. We further conclude that, in exercising that authority, the Agency did not act arbitrarily or capriciously in finding that ADEC’s BACT decision in this instance lacked evidentiary support. EPA’s orders, therefore, were neither arbitrary nor capricious. The judgment of the Court of Appeals is accordingly

Affirmed.

Justice Kennedy,

with whom The Chief Justice, Justice Scalia, and Justice Thomas join, dissenting. ‘

The majority, in my respectful view, rests its holding on mistaken premises, for its reasoning conflicts with the express language of the Clean Air Act (CAA or Act), with sound rules of administrative law, and with principles that preserve the integrity of States in our federal system. The State of Alaska had in place procedures that were in full compliance with the governing statute and accompanying regulations promulgated by the Environmental Protection Agency (EPA). As I understand the opinion of the Court and the parties’ submissions, there is no disagreement on this point. Alaska followed these procedures to determine the best available control technology (BACT). EPA, how*503ever, sought to overturn the State’s decision, not by the process of judicial review, but by administrative fiat. The Court errs, in my judgment, by failing to hold that EPA, based on nothing more than its substantive disagreement with the State’s discretionary judgment, exceeded its powers in setting aside Alaska’s BACT determination.

I

As the majority explains, the case begins with §§ 113(a)(5) and 167 of the Act. 42 U. S. C. §§ 7413(a)(5), 7477. These provisions give EPA authority to enforce “requirements” of the CAA. The meaning of the word “requirement,” though, is not defined in these provisions. Other provisions of the Act must be consulted. All parties agree that the requirement in this case is the “preconstruction requirement” that a “major emitting facility” be “subject to the best available control technology [BACT] for each pollutant subject to regulation under this chapter emitted from, or which results from, such facility.” § 7475(a)(4). BACT, in turn, is defined as

“an emission limitation based on the maximum degree of reduction of each pollutant subject to regulation under this chapter emitted from or which results from any major emitting facility, which the permitting authority, on a ease-by-case basis, taking into account energy, environmental, and economic impacts and other costs, determines is achievable for such facility through application of production processes and available methods, systems, and techniques . . . .” §7479(3).

The majority holds that, under the CAA, state agencies are vested with “initial responsibility for identifying BACT in line with the Act’s definition of that term” and that EPA has a “broad oversight role” to ensure that a State’s BACT determination is “reasonably moored to the Act’s provisions.” Ante, at 484-485. The statute, however, contem*504plates no such arrangement. It directs the “permitting authority” — here, the Alaska Department of Environmental Conservation (ADEC) — to “determine” what constitutes BACT. To “determine” is not simply to make an initial recommendation that can later be overturned. It is “[t]o decide or settle . . . conclusively and authoritatively.” American Heritage Dictionary 495 (4th ed. 2000). Cf. 5 U. S. C. § 554 (“to be determined on the record after opportunity for an agency hearing”).

The BACT definition presumes that the permitting authority will exercise discretion. It presumes, in addition, that the BACT decision will accord full consideration to the statutory factors and other relevant and necessary criteria. Contrary to the majority’s holding, the statute does not direct the State to find as BACT the technology that results in the “maximum reduction of a pollutant achievable for [a] facility” in the abstract. Ante, at 485 (internal quotation marks omitted). Indeed, for a State to do so without regard to the other mandatory criteria would be to ignore the words of the statute. The Act requires a more comprehensive judgment. It provides that the permitting authority must “tak[e] into account” a set of contextual considerations — “energy, environmental, and economic impacts and other costs” — to identify the best control technology “on a case-by-case basis.” 42 U. S. C. § 7479(3). The majority reaches its narrow view of the scope of the State’s discretion only by wresting two adjectives, “maximum” and “achievable,” out of context. In doing so, it ignores “the cardinal rule that a statute is to be read as a whole.” King v. St. Vincent’s Hospital, 502 U. S. 215, 221 (1991):

To be sure, §§ 113(a)(5) and 167 authorize EPA to enforce requirements of the Act. These provisions, however, do not limit the States’ latitude and responsibility to balance all the statutory factors in making their discretionary judgments. If a State has complied with the Act’s requirements, §§ 113(a)(5) and 167 are not implicated and can supply no sep*505arate basis for EPA to exercise a supervisory role over a State’s discretionary decision. The Court of Appeals for the Ninth Circuit had it altogether backwards when it reasoned that, “because neither Section 113(a)(5) nor Section 167 contains any exemption for requirements that involve the state’s exercise of discretion,” EPA had the authority to issue orders countermanding the State’s BACT determination. 298 F. 3d 814, 820 (2002). The question is not whether the two sections contain any exemption. Rather, it is about the nature of the Act’s requirements and whether EPA has the authority to set aside a BACT determination when no requirement of the Act was violated in the first place. In affirming the judgment of the Court of Appeals, the majority repeats the same analytical error. See ante, at 490 (“We fail to see why Congress, having expressly endorsed an expansive surveillance role for EPA in two. independent CAA provisions, would then implicitly preclude the Agency from verifying substantive compliance with [BACT]...”). When the statute is read as a whole, it is clear that the CAA commits BACT determinations to the discretion of the relevant permitting authorities. Unless an objecting party, including EPA, prevails on judicial review, the determinations are conclusive.

Here the state agency, ADEC, recognized it was required to make a BACT determination. It issued two detailed reports in response to comments by interested parties and concluded that Low Nitrogen Oxide (NOx) was BACT. The requirement that the agency weigh the list of statutory factors, study all other relevant considerations, and decide the technology that can best reduce pollution within practical constraints was met in full. As even EPA acknowledged, ADEC “provided] a detailed accounting of the process.” App. 286. This is not a ease, then, where the state agency failed to have a BACT review procedure in place or altogether refused to apply the statute’s formal requirements. EPA’s only quarrel is with ADEC’s substantive conclusion. *506In disagreeing with ADEC, EPA’s sole contention, in the section of its order titled “Findings of Fact,” is that “[selective catalytic reduction] is BACT.” App. to Pet. for Cert. 30a, 34a (emphasis added). In addition, EPA does not allege that using Low NOx would violate other CAA requirements, such as the National Ambient Air Quality Standards, Alaska’s Prevention of Significant Deterioration (PSD) increments, or other applicable emission standards, see 42 U. S. C. § 7475(a)(3). On this state of the record there is no deviation from any statutory “requirement.” As a result, EPA has no statutory basis to invoke the enforcement authority of .§§ 113(a)(5) and 167.

When Congress intends to give EPA general supervisory authority, it says so in clear terms. In addition to requiring EPA’s advance approval of BACT determinations in some instances, 42 U. S. C. § 7475(a)(8), the statute grants EPA powers to block the construction or operation of polluting sources in circumstances not at issue here, §§ 7426(b), (c)(1), 7410(a)(2)(D)(i). Outside the context of the CAA, Congress likewise knows how to establish federal oversight in unambiguous language. See, e. g., 42 U. S. C. § 1396a(a)(13)(A) (1994 ed.) (requiring, under the Medicaid Act, reimbursement according to rates that a “State finds, and makes assurances satisfactory to the Secretary [of Health and Human Services], are reasonable and adequate to meet the costs which must be incurred by efficiently and economically operated facilities”); Wilder v. Virginia Hospital Assn., 496 U. S. 498 (1990). No analogous language is used in the statutory definition of BACT.

EPA insists it needs oversight authority to prevent a “race to the bottom,” where jurisdictions compete with each other to lower environmental standards to attract new industries and keep existing businesses within their borders. Whatever the merits of these arguments as a general matter, EPA’s distrust of state agencies is inconsistent with the Act’s clear mandate that States bear the primary role in control*507ling pollution and, here, the exclusive role in making BACT determinations. In “cho[osingj not to dictate a Federal response to balancing sometimes conflicting goals” at the expense of “[m]aximum flexibility and State discretion,” H. R. Rep. No. 95-294, p. 146 (1977), Congress made the overriding judgment that States are more responsive to local conditions and can strike the right balance between preserving environmental quality and advancing competing objectives. By assigning certain functions to the States, Congress assumed they would have a stake in implementing the environmental objectives of the Act. At the same time, Congress charged EPA with setting ambient standards and enforcing emission. limits, 42 U. S. C. § 7475(a)(3), to ensure that the Nation takes the necessary steps to reduce air pollution.

The presumption that state agencies are not to be trusted to do their part is unwarranted in another respect: EPA itself said so. As EPA concedes, States, by and large, take their statutory responsibility seriously, and EPA sees no reason to intervene in the vast majority of cases. Brief for Respondents 30, n. 9; 57 Fed. Reg. 28095 (1992) (“States have been largely successful in [‘administering and enforcing the various components of the PSD program’], and EPA’s involvement in interpretative and enforcement issues is limited . . .”). In light of this concession, EPA and amici not only fail to overcome the established presumption that States act in good faith, see Alden v. Maine, 527 U. S. 706, 755 (1999) (“We are unwilling to assume the States will refuse to honor ... or obey the binding laws of the United States”), but also admit that their fears about a race to the bottom bear little relation to the real-world experience under the statute. See ante, at 502 (“We see no reason not to take EPA at its word”).

II

The statute contains safeguards to correct arbitrary and capricious BACT decisions when they do occur. Before EPA approves a State’s PSD permit program that allows a *508state agency to make BACT determinations, EPA must be satisfied that the State provides “an opportunity for state judicial review.” 61 Fed. Reg. 1882 (1996). Furthermore, before an individual permit may issue, the State must allow all “interested persons,” including “representatives of the [EPA] Administrator,” to submit comments on, among other things, “control technology requirements.” 42 U. S. C. § 7475(a)(2). To facilitate EPA’s participation in the State’s public comment process, the statute further provides that specific procedures be followed to inform the EPA Administrator of “every action” taken in the course of the permit approval process. § 7475(d) (“Each State shall transmit to the Administrator a copy of each permit application relating to a major emitting facility received by such State and provide notice to the Administrator of every action related to the consideration of such permit”). Any person who participated in the comment process can pursue an administrative appeal of the State’s decision, followed, as mentioned, by judicial review in state courts.

EPA followed none of the normal procedures here. Only after the period for public comments expired did it intervene and seek to overturn Alaska’s decision that Low NOx was BACT. To justify its decision to opt out of the State’s administrative and judicial review process and, instead, to issue a unilateral order after everyone had spoken, EPA complains that it has not before intervened in “any State administrative review proceedings in State courts” and should not now be forced to do so. Tr. of Oral Arg. 35. With scant analysis, the majority agrees. Ante, at 492 (“It would be unusual, to say the least, for Congress to remit a federal agency enforcing federal law solely to state court. We decline to read such an uncommon regime into the Act’s silence”). The problem, of course, is that it is all the more unusual to allow a federal agency to take unilateral action to set aside a State’s administrative decision.

*509Despite EPA’s protestations, the statute makes explicit provision for EPA to challenge a state agency’s BACT determination in state proceedings. The statute requires States to set up an administrative process for “interested persons” to submit comments. § 7475(a)(2). “[Interested persons,” Congress took care to note, include “representatives of the [EPA] Administrator.” Ibid.; see also Alaska Stat. §46.14.990(20) (2002) (defining “person” to include “an agency of the United States”). Given that EPA itself requires, as a condition of approving a State’s PSD program, that this process culminate in judicial review in state courts, 61 Fed. Reg., at 1882, it follows that EPA, a subset of all “interested persons,” must take the same procedural steps and cannot evade the more painstaking state process by a mere stroke of the pen under the agency’s letterhead.

On a more fundamental level, EPA and the majority confuse a substantive environmental statute like the CAA with a general administrative law statute like the Administrative Procedure Act (APA). EPA, the federal agency charged only with the CAA’s implementation, has no roving commission to ferret out arbitrary and capricious conduct by state agencies under the state equivalent of the APA. That task is left to state courts. See Idaho v. Coeur d’Alene Tribe of Idaho, 521 U. S. 261, 276 (1997) (“[T]he elaboration of administrative law ... is one of the primary responsibilities of the state judiciary”).

Like federal courts, state courts are charged with reviewing agency actions to ensure that they comport with principles of rationality and due process. See, e.g., 5 U. S. C. § 706(2)(A); Alaska Stat. § 44.62.570(b)(3) (2002). Counsel for respondents were unable to identify, either in their briefs or at oral argument, a single State that “does not have in its law the requirement that its own agencies ... act rationally.” Tr. of Oral Arg. 30. Although it remains an open question whether EPA can bypass the state judiciary and go directly into federal district court under 28 U. S. C. § 1345, the avail*510ability of state judicial review defeats the Government’s argument that, absent EPA’s oversight, there is a legal vacuum where BACT decisions are not subject to review.

Requiring EPA to seek administrative and judicial review of a State’s BACT determination, instead of allowing it to be overturned by fiat, avoids the anomaly of shifting the burden of pleading and of initiating litigation from EPA to the State. Whether the BACT decision is reviewed in state court, or in federal district court if that option is available, see supra, at 509 and this page, EPA, as petitioner, bears the initial burden and costs of filing a petition for review alleging that the State acted arbitrarily. Under the scheme endorsed by the majority today, the tables are turned. Once EPA has issued an enforcement order, and the State seeks to invalidate that order, the State bears the burden of alleging that EPA acted arbitrarily. EPA and the majority concede that, because States enjoy substantial discretion in making BACT determinations, courts reviewing EPA’s order must ask not simply whether EPA acted arbitrarily but the convoluted question whether EPA acted arbitrarily in finding the State acted arbitrarily. Even under this unwieldy standard of review, and even if the burdens of persuasion and production remain with EPA, see ante, at 493-494, the initial burden of pleading and litigation now belongs to the State.

To make its decision more palatable, the majority holds that EPA still bears the burdens of production and persuasion, but there is little authority for this. The Court purports to rely on McCormick on Evidence for the proposition that “ ‘looking for the burden of pleading is not a foolproof guide to the allocation of the burdens of proof.’” Ante, at 494, n. 17 (quoting 2 J. Strong, McCormick on Evidence § 337, pp. 411-412 (5th ed. 1999)). The' example — affirmative defense — discussed in that passage of the treatise, however, is far afield from the issues raised in this case. In fact, the treatise instructs that “[i]n most cases, the party who has the burden of pleading a fact will have the burdens of producing *511evidence and of persuading the jury of its existence as well.” Id., at 411. This is because “[t]he burdens of pleading and proof with regard to most facts have been and should be assigned to the plaintiff who generally seeks to change the present state of affairs and who therefore naturally should be expected to bear the risk of failure of proof or persuasion.” Id., at 412. In this case, EPA changed the status quo ante by issuing an order invalidating ADEC’s decision. Without upsetting accepted evidentiary principles, the majority cannot explain why EPA, as respondent in federal court — as opposed to the State, as petitioner alleging that EPA’s/aii accompli was arbitrary — should bear the burdens of persuasion and production, or how this unusual reallocation of burdens should work in practice.

In any event, even the majority accepts that, under its reading of the statute, the State now bears the burden of pleading. With this burden-shifting benefit alone, EPA is most unlikely to follow the procedure, prescribed by federal law, of participating in the State’s administrative process and seeking judicial review in state courts. Instead, EPA can simply issue a unilateral order invalidating the State’s BACT determination and put the burden on the State to challenge EPA’s order. This end run around the State’s process is sure to undermine it. Unless Congress was on a fool’s errand, the loophole the majority finds goes only to demonstrate the inconsistency between its approach and the statutory scheme.

There is a further, and serious, flaw in the Court’s ruling. Suppose, before EPA issued its orders setting aside the State’s BACT determination, an Alaska state court had reviewed the matter and found no error of law or abuse of discretion in ADEC’s determination. The majority’s interpretation of the statute would allow EPA to intervene at this point for the first time, announce that ADEC’s determination is unreasoned under the CAA, and issue its own orders nullifying the state court’s ruling. This reworking of the bal*512ance between State and Federal Governments, not to mention the reallocation of authority between the Executive and Judicial Branches, shows the implausibility of the majority’s reasoning.

If a federal agency were to exercise an analogous power to review the decisions of federal courts, the arrangement would violate the well-established rule that the judgments of Article III courts cannot be revised by the Executive or Legislative Branches. See Hayburn’s Case, 2 Dall. 409, 410, n. (1792) (“ ‘[B]y the Constitution, neither the Secretary [of] War, nor any other Executive officer, nor even the Legislature, are authorized to sit as a court of errors on .. . judicial acts or opinions . . see also Plaut v. Spendthrift Farm, Inc., 514 U. S. 211 (1995). The principle that judicial decisions cannot be reopened at the whim of the Executive or the Legislature is essential to preserving separation of powers and judicial independence. Judges cannot, without sacrificing the autonomy of their office, put onto the scales of justice some predictive judgment about the probability that an administrator might reverse their rulings.

The Court today denies state judicial systems the same judicial independence it has long guarded for itself — only that the injury here is worse. Under the majority’s holding, decisions by state courts would be subject to being overturned, not just by any agency, but by an agency established by a different sovereign. We should be reluctant to interpret a congressional statute to deny to States the judicial independence guaranteed by their own constitutions. See Buckalew v. Holloway, 604 P. 2d 240, 245 (Alaska 1979) ("There is no doubt that judicial independence was a paramount concern of the delegates [to the Alaska Constitutional Convention]”);' see also, e. g., Cal. Const., Art. 3, § 3 ("The powers of state government are legislative, executive, and judicial. Persons charged with the exercise of one power may not exercise either of the others except as permitted by this Constitution”); see also 7 B. Witkin, Summary of *513California Law § 107, pp. 159-160 (9th ed. 1988) (“[Under] the principle of separation of powers ... , one [department] cannot exercise or interfere with the functions of either of the others”). The Federal Government is free, within its vast legislative authority, to impose federal standards. For States to have a role, however, their own governing processes must be respected. New York v. United States, 505 U. S. 144 (1992). If, by some course of reasoning, state courts must live with the insult that their judgments can be revised by a federal agency, the Court should at least insist upon a clear instruction from Congress. That directive cannot be found here. Cf. Gregory v. Ashcroft, 501 U. S. 452, 460 (1991) (“[I]f Congress intends to alter the usual constitutional balance between the States and the Federal Government, it must make its intention to do so unmistakably clear in the language of the statute” (internal quotation marks omitted)).

There is a final deficiency in the scheme the majority finds in the statute. Nothing in the Court’s analysis prevents EPA from issuing an order setting aside a BACT determination months, or even years, later. Congress cannot have intended this result. After all, when Congress provides for EPA’s involvement, it directs the agency to act sooner rather than later by establishing a preauthorization procedure. 42 U. S. C. § 7475(a)(8). The majority misses the point when it faults ADEC for “overlooking] the obvious difference between a statutory requirement... and a statutory authorization.” Ante, at 491 (emphasis deleted). ADEC does not overlook the difference between approval before the fact and oversight after the fact. Rather, ADEC, unlike the majority, recognizes that the Act’s explicit provision for a preauthorization process underscores the need for finality in state permitting decisions, making implausible an interpretation of the statute that would allow a post hoc veto procedure that upsets the same reliance and expectation interests.

*514The majority’s initial response that “[t]his case threatens no such development [because] [i]t involves preconstruction orders issued by EPA . . . , not postconstruction federal Agency directives,” ante, at 495, provides no assurance that the logic of its reasoning would not in the future allow EPA’s belated interventions. When the majority confronts the problem, it concludes that “EPA, we are confident, could not indulge in the inequitable conduct ADEC and the dissent hypothesize while the federal courts sit to review EPA’s actions.” Ibid. The authority it cites for this proposition, however, consists of nothing more than a religious exemption case that is far removed from the issues presented here and a dissent from a case that has been overruled in part. Ibid. State agencies rely on this dictum at their own risk.

The majority’s reassurance to the States will likely be to no avail. “The principle that the United States are not bound by any statute of limitations, nor barred by any laches of their officers, however gross, in a suit brought by them as a sovereign Government to enforce a public right, or to assert a public interest, is established past all controversy or doubt.” United States v. Beebe, 127 U. S. 338, 344 (1888); see also United States v. Summerlin, 310 U. S. 414, 416 (1940) (“It is well settled that the United States is not bound by state statutes of limitation or subject to the defense of laches in enforcing its rights”); Utah Power & Light Co. v. United States, 243 U. S. 389, 409 (1917) (“[L]aches or neglect of duty on the part of officers of the Government is no defense to a suit by it to enforce a public right or protect a public interest. ... A suit by the United States to enforce and maintain its policy . . . stands upon a different plane in this and some other respects from the ordinary private suit . . .”). Section 167, moreover, is mandatory. Once a violation of a statutory “requirement” is found, “[t]he Administrator shall . . . take such measures, including issuance of an order, or seeking injunctive relief, as necessary to prevent the construction or modification of a major emitting fa*515cility which does not conform to the requirements of this part...42 U. S. C. §7477. In short, EPA’s enforcement authority can — indeed, must — be exercised at any point. In light of our precedents a court would be hard pressed to hold otherwise.

The majority seeks to limit the consequence of its holding by quoting the response by respondents’ counsel at oral argument that ADEC could “absolutely” arrive at the same BACT determination if only it would pile on another layer of procedure and justify its decision on an “‘appropriate record.’” Ante, at 501-502 (quoting Tr. of Oral Arg. 35). As the Court of Appeals recognized in a prior case, however, this option gives no solace to the States:

“The hardship is the process itself. Process costs money. If a federal licensee must spend years attempting to satisfy an elaborate, shifting array of state procedural requirements, then he must borrow a fortune to pay lawyers, economists, accountants, archaeologists, historians, engineers, recreational consultants, environmental consultants, biologists and others, with no revenue, no near-term prospect of revenue, and no certainty that there ever will be revenue. Meanwhile, politics, laws, interest rates, construction costs, and costs of alternatives change. Undue process may impose cost and uncertainty sufficient to thwart the federal determination that a power project should proceed.” Sayles Hydro Associates v. Maughan, 985 F. 2d 451, 454 (CA9 1993).

If there is to be a second look, notwithstanding the 18 months ADEC spent analyzing BACT, a third or fourth look is just as permissible. The majority creates a sort of Zeno’s paradox for state agencies. Because there can always be an additional procedure to ensure that the preceding process was followed, no matter how many steps States take toward the objective, they may never reach it.

*516This is a most regrettable result. In the proper discharge of their responsibilities to implement the CAA in different conditions and localities nationwide, the States maintain permanent staffs within special agencies. These state employees, who no doubt take pride in their own resourcefulness, expertise, and commitment to the law, are the officials directed by Congress to make case-by-case, site-specific, determinations under the Act. Regulated persons and entities should be able to consult an agency staff with certainty and confidence, giving due consideration to agency recommendations and guidance. After today’s decision, however, a state agency can no longer represent itself as the real governing body. No matter how much time was spent in consultation and negotiation, a single federal administrator can in the end set all aside by a unilateral order. This is a great step backward in Congress’ design to grant States a significant stake in developing and enforcing national environmental objectives.

If EPA were to announce that permit applications subject to BACT review must be submitted to it in the first instance and can be forwarded to the State only with EPA’s advance approval, I should assume even the majority would find the basic structure of the BACT provisions undercut. In practical terms, however, the majority displaces state agencies, and degrades their role, in much the same way. In the case before us the applicant made elaborate submissions to ADEC. For over a year and a half, there ensued the constructive discourse that is the very object of the agency process, with both the ADEC staff and the applicant believing the State’s decision would be dispositive. EPA did not participate in the administrative process, but waited until after the record was closed to intervene by issuing an order setting aside the BACT determination.

We are advised that an applicant sometimes must spend up to $500,000 on the permit process and that, for a complex project, the time for approval can take from five to *517seven years. Brief for National Environmental Development Association et al. as Amici Curiae 8. Under the new multiple-tiered process, permit expenditures become less justified, state officials less credible, reliance less certain. The Court should be under no illusion that its decision respects the State’s administrative process.

The federal balance is remitted, in many instances, to Congress. Here the Court remits it to a single agency official. This is inconsistent with the assurance Congress gave to regulated entities when it allowed state agencies to decide upon the grant or denial of a permit under the BACT provisions of the CAA.

Ill

In the end EPA appears to realize the weakness of its arguments and asks us simply to defer to its expertise in light of the purported statutory ambiguity. See Brief for Respondents 41-43 (asking for deference under Chevron U. S. A. Inc. v. Natural Resources Defense Council, Inc., 467 U. S. 837 (1984)). To its credit, the majority holds Chevron deference inapplicable. Deference is inappropriate for all the reasons the majority recites, ante, at 487-488, plus one more: The statute is not in any way ambiguous. As a result, our inquiry should proceed no further.

Actions, however, speak louder than words, and the majority ends up giving EPA the very Chevron deference — and more — it says should be denied. The Court’s opinion is chock full of Chevron-like language. Compare 467 U. S., at 843 (“whether the agency’s answer is based on a permissible construction of the statute”); id., at 845 (“whether the Administrator’s view ... is a reasonable one”), with ante, at 488 (“[EPA’s] arguments do not persuade us to reject [them] as impermissible”); ante, at 493 (“That rational interpretation, we agree, is surely permissible”). So deficient are its statutory arguments that the majority must hide behind Chevron’s vocabulary, despite its explicit holding that Chevron does not apply. In applying Chevron de facto under these *518circumstances, however, the majority undermines the well-established distinction our precedents draw between Chevron and less deferential forms of judicial review.

The broader implication of today’s decision is more unfortunate still. The CAA is not the only statute that relies on a close and equal partnership between federal and state authorities to accomplish congressional objectives. See, e. g., New York v. United States, 505 U. S., at 167 (listing examples). Under the majority’s reasoning, these other statutes, too, could be said to confer on federal agencies ultimate decisionmaking authority, relegating States to the role of mere provinces or political corporations, instead of coequal sovereigns entitled to the same dignity and respect. Cf. Alden v. Maine, 527 U. S. 706 (1999). If cooperative federalism, Hodel v. Virginia Surface Mining & Reclamation Assn., Inc., 452 U. S. 264, 289 (1981), is to achieve Congress’ goal of allowing state governments to be accountable to the democratic process in implementing environmental policies, federal agencies cannot consign States to the ministerial tasks of information gathering and making initial recommendations, while reserving to themselves the authority to make final judgments under the guise of surveillance and oversight.

For these reasons, and with all respect, I dissent from the opinion and the judgment of the Court.

CUDAHY, Circuit Judge.

The Petitioner, Wisconsin Electric Power Company (“WEPCO”), challenges two final determinations issued by the Environmental Protection Agency (the “EPA”). In these determinations, the EPA concluded that WEPCO’s proposed renovations to its Port Washington power plant would subject the plant to certain pollution control provisions of the Clean Air Act, as amended, 42 U.S.C. §§ 7401 et seq. (1982). We affirm in part, vacate in part and remand to the EPA.

I. The Underlying Dispute

A. Relevant Provisions of the Clean Air Act

Some discussion of the Clean Air Act is required before turning to the merits of this case. In 1970, Congress enacted the Clean Air Act Amendments, Pub.L. No. 91-604, 84 Stat. 1676, to establish minimum air quality standards that would regulate the emission of certain pollutants into the atmosphere. To this end, Congress instructed the EPA to develop National Ambient Air Quality Standards (“NAAQS”) that would specify the maximum permissible concentration of air pollutants in different areas across the country.

In section 111 of the 1970 Amendments, Congress required the EPA to promulgate New Source Performance Standards (“NSPS”) in order to regulate the emission of air pollutants from new sources. These standards addressed hourly rates of emission and, in addition to new sources, applied to modifications of existing facilities that created new or increased pollution. Indeed, section 111(a)(2) of the Act stated that NSPS would apply to

any stationary source, the construction or modification of which is commenced after the publication of regulations (or, if earlier, proposed regulations) prescribing a standard of performance under this section which will be applicable to such source.

42 U.S.C. § 7411(a)(2) (emphasis supplied). Congress then defined “modification” as

any physical change in, or change in the method of operation of, a stationary source which increases the amount of any air pollutant emitted by such source or which results in the emission of any air pollutant not previously emitted.

42 U.S.C. § 7411(a)(4) (emphasis supplied).

Subsequently, faced with only varying degrees of success in controlling pollution in different parts of the country, Congress enacted the Clean Air Act Amendments of 1977, Pub.L. No. 95-95, 91 Stat. 685 (codified at 42 U.S.C. §§ 7401-7642 (1982)). Congress revised the NSPS so that regulated sources of pollution would have to use “the best system of continuous emission reduction which (taking into consideration the costs of achieving such emission reduction, and any nonair quality health and environmental impact and energy requirements) the Administrator determines has been adequately demonstrated....” 42 U.S.C. § 7411(a)(1)(C). In addition, Congress added a program for the Prevention of Significant Deterioration (“PSD”), concerned with increases in total annual emissions, to ensure that operators of regulated sources in relatively unpolluted areas would not allow a decline of air quality to the minimum level permitted by NAAQS. Air quality is preserved in this program by *905requiring sources to limit their emissions to a "baseline rate”; regulated owners or operators in areas that have attained NAAQS must obtain a permit before constructing or modifying facilities. 42 U.S.C. § 7475(a)(1). Congress also essentially adopted its NSPS definition of “modification” for the PSD program. 42 U.S.C. § 7479(2)(C).

From this statutory framework, the EPA promulgated regulations for both the NSPS and PSD programs. In this case, its regulations concerning modifications are central. The EPA defines “modification” in substantially the same terms used by Congress:

[A]ny physical or operational change to an existing facility which results in an increase in the emission rate to the atmosphere of any pollutant to which a standard applies shall be considered a modification within the meaning of section 111 [42 U.S.C. § 7411] of the Act.

40 C.F.R. § 60.14(a) (1988). To determine whether a physical change constitutes a modification for purposes of NSPS, the EPA must determine whether the change increases the facility’s hourly rate of emission. 40 C.F.R. § 60.14 (1988). For PSD purposes, current EPA regulations provide that an increase in the total amount of emissions activates the modification provisions of the regulations. 40 C.F.R. § 52.21(b)(3) (1988).

Even at first blush, the potential reach of these modification provisions is apparent: the most trivial activities — the replacement of leaky pipes, for example — may trigger the modification provisions if the change results in an increase in the emissions of a facility. As a result, the EPA promulgated specific exceptions to the modification provisions:

The following shall not, by themselves, be considered modifications under this part:

(1)Maintenance, repair, and replacement which the Administrator determines to be routine for a source category

(2) An increase in production rate of an existing facility, if that increase can be accomplished without a capital expenditure on that facility.

(3) An increase in the hours of operation ....

40 C.F.R. § 60.14(e) (1988) (NSPS program); see 40 C.F.R. § 52.21(b)(2)(iii) (1988) (PSD program). These regulations (and the statutes from which they derive) are the focal point of this case.

B. WEPCO’s Proposed Life-Extension Project

WEPCO’s Port Washington electric power plant is located on Lake Michigan north of Milwaukee, Wisconsin. The plant consists of five coal-fired steam generating units that were placed in operation between 1935 and 1950. Each generating unit has a design capacity of 80 megawatts, but the recent performance of some of the units has declined due to age-related deterioration of the physical plant.

WEPCO and its consultant, Bechtel Eastern Power Corporation, conducted a Plant Availability Study in 1983 to examine and assess the condition of the power plant. As a result of the Study, WEPCO concluded “that extensive renovation of the five units and the plant common facilities is needed if operation of the plant is to be continued.” Letter from Thomas J. Cassi-dy, Executive Vice President of WEPCO, to Jacqueline K. Reynolds, Secretary to the Public Service Commission of Wisconsin, at 2 (July 8, 1987) [Cassidy Letter] (emphasis supplied). The Study noted that the air heaters on the first four units had deteriorated severely, while the rear steam drums in units 2 through 5 had experienced serious cracking.1 Air heater deterioration prevented units 1 and 4 from operating at full capacity, while the potential for steam drum blowout required a reduction in pressure (and output) in units 2 and 3. The possibility of catastrophic failure (steam *906drum blowout) in unit 5 was so great that WEPCO shut down the unit completely.

As a result of this Study, WEPCO submitted a proposed replacement program (which it termed a “life extension” project) to the Wisconsin Public Service Commission for its approval, as required by state law. Wis.Stat. § 196.49 (1987). WEPCO explained in its proposal that “[rjenovation is necessary to allow the Port Washington units to operate beyond their currently planned retirement dates of 1992 (units 1 and 2) and 1999 (units 3, 4 and 5) ... [and that renovation would render the plant] capable of generating at its designed capability until year 2010.... ” Cassidy Letter at 1-2. Among the renovations required were repair and replacement of the turbine-generators, boilers, mechanical and electrical auxiliaries and the common plant support facilities. Id. at 1. After preliminary review of the program, the Public Service Commission consulted the Wisconsin Department of Natural Resources (which then consulted EPA Region V) to determine whether WEPCO needed to obtain a PSD permit before commencing the repair and replacement program. David Kee, the Director of EPA Region V’s Air and Radiation Division, then referred the matter to EPA Headquarters. See, e.g., 40 C.P.R. § 60.5 (1988) (discussing the EPA’s procedures regarding determinations of construction or modification).

EPA staff members conferred with WEPCO representatives between March and September 1988 to gain additional information regarding the proposed repair and replacement project. On September 9, 1988, EPA Acting Assistant Administrator Don R. Clay issued a memorandum in which he preliminarily concluded that the project would subject the plant to both NSPS and PSD requirements. Memorandum from Don R. Clay, Acting Assistant Administrator for Air and Radiation of the EPA, to David A. Kee, Director of Air and Radiation Division, Region V (Sept. 9, 1988) [Clay Memorandum]. The Clay Memorandum pointed out that the project would constitute a “physical change” resulting in an increase of production and emissions, which would therefore subject the plant to the relevant strictures of the Clean Air Act. Id. at 3-4. Further, the Clay Memorandum dismissed WEPCO’s contention that the program was routine and was therefore exempt from the requirements of NSPS and PSD. This conclusion was adopted in to to by EPA Administrator Lee M. Thomas. Letter from Lee M. Thomas, Administrator of the EPA, to John Boston, Vice President of WEPCO (Oct. 14, 1988) [Thomas Letter],

Following the Thomas Letter, WEPCO continued to conduct capacity tests on the units. Based upon these tests, Assistant Administrator Clay issued a “revised final determination” that generally affirmed the EPA’s earlier findings, but modified the baseline figures used by the EPA for units 2 and 3. Letter from Don R. Clay, Acting Assistant Administrator for Air and Radiation of the EPA, to John W. Boston, Vice President of WEPCO (Feb. 15, 1989) [Supplemental Determination],

Alleging that the EPA has misconstrued both the Clean Air Act and its own regulations, WEPCO appeals the EPA’s final determination. We have jurisdiction to hear this appeal pursuant to 42 U.S.C. § 7607(b) (1982).

II. Standard of Review

Courts have generally accorded substantial deference to the EPA’s interpretation of the Clean Air Act Amendments, reasoning that “considerable weight should be accorded to an executive department’s construction of a statutory scheme it is entrusted to administer....” Chevron U.S.A. Inc. v. Natural Resources Defense Council, Inc., 467 U.S. 837, 844, 104 S.Ct. 2778, 2782, 81 L.Ed.2d 694 (1984); see Union Elec. Co. v. EPA, 427 U.S. 246, 256, 96 S.Ct. 2518, 2525, 49 L.Ed.2d 474 (1976); Train v. Natural Resources Defense Council, Inc., 421 U.S. 60, 75, 87, 95 S.Ct. 1470, 1479, 1485, 43 L.Ed.2d 731 (1975); ASARCO Inc. v. EPA, 578 F.2d 319, 325 (D.C.Cir.1978). This deference with regard to the Clean Air Act follows logically from the highly technical provisions of the Amendments, Chevron, 467 U.S. at 848, 104 S.Ct. at 2784, and is consistent with the *907Administrative Procedure Act, which provides that agency actions are to be set aside only if they are “arbitrary, capricious, an abuse of discretion, or otherwise not in accordance with law.” 5 U.S.C. § 706(2).

To be sure, this standard does not give the EPA unbridled discretion to construe the Clean Air Act Amendments free from judicial oversight. We must consider whether the EPA’s construction comports with its statutory mandate and Congress’s intent in enacting clean air legislation. But we cannot simply substitute our judgment for that of the EPA. Our role has been sharply defined and limited by the Supreme Court:

When a court reviews an agency’s construction of the statute which it administers, it is confronted with two questions. First, always, is the question whether Congress has directly spoken to the precise question at issue. If the intent of Congress is clear, that is the end of the matter; for the court, as well as the agency, must give effect to the unambiguously expressed intent of Congress. If, however, the court determines Congress has not directly addressed the precise question at issue, the court does not simply impose its own construction on the statute, as would be necessary in the absence of an administrative interpretation. Rather, if the statute is silent or ambiguous with respect to the specific issue, the question for the court is whether the agency’s answer is based on a permissible construction of the statute.

Chevron, 467 U.S. at 842-43, 104 S.Ct. at 2781-82 (footnotes omitted).

Further, we defer even more to an agency’s construction of its own regulations. Lyng v. Payne, 476 U.S. 926, 939, 106 S.Ct. 2333, 2341, 90 L.Ed.2d 921 (1986); see Wilkins v. Sullivan, 889 F.2d 135, 139 (7th Cir.1989); Homemakers North Shore, Inc. v. Bowen, 832 F.2d 408, 411 (7th Cir. 1987) (agency construction of its regulations usually upheld). An agency’s interpretation must be upheld “ ‘unless it is plainly erroneous or inconsistent with the regulation.’ ” Udall v. Tallman, 380 U.S. 1, 16-17, 85 S.Ct. 792, 801-802, 13 L.Ed.2d 616 (1965) (quoting, in part, Bowles v. Seminole Rock & Sand Co., 325 U.S. 410, 413-14, 65 S.Ct. 1215, 1217, 89 L.Ed. 1700 (1945)). The principle of deference has particular force where, as is the case here, the subject being regulated is technical and complex. Aluminum Co. of Am. v. Central Lincoln Peoples’ Util. Dist., 467 U.S. 380, 390, 104 S.Ct. 2472, 2479, 81 L.Ed.2d 301 (1984); Wilkins v. Sullivan, 889 F.2d at 140; see also Skidmore v. Swift & Co., 323 U.S. 134, 140, 65 S.Ct. 161, 164, 89 L.Ed. 124 (1944) (rulings of agency constitute bodies of experience and informed judgment).

III. Like-Kind Replacement and Modification under the Act

A. The Underlying Statutory Framework

With these principles in mind, we may address the merits. We must first consider whether WEPCO’s Port Washington replacement program constitutes a modification under the terms of the controlling statute, 42 U.S.C. section 7411(a)(4). Cf. Blue Chip Stamps v. Manor Drug Stores, 421 U.S. 723, 756, 95 S.Ct. 1917, 1935, 44 L.Ed.2d 539 (1975) (Powell, J., concurring) (“The starting point in every case involving construction of a statute is the language itself.”). Section 7411(a)(4) defines modification as “any physical change ... which increases the amount of any air pollutant emitted_” 42 U.S.C. § 7411(a)(4). Both parts of this definition — any physical change and an increase in emissions — must be satisfied before a replacement will be considered a “modification.”

1. Physical Change

Certainly, under the plain terms of the Act, WEPCO’s replacement program constitutes a “physical change.” WEPCO proposes to replace rear steam drums on units 2, 3, 4 and 5; each of these steam drums measures 60 feet in length, 50.5 inches in diameter and 5.25 inches in thickness. Clay Memorandum at 4. In addition, WEPCO plans to replace another ma*908jor component, the air heaters, in units 1-4. To implement this four-year program, WEPCO will need to make the replacements by taking the units successively out of service for nine-month periods. Id. These steps clearly amount to a “physical change” in the Port Washington plant. See Butler, New Source Netting in Nonat-tainment Areas under the Clean Air Act, 11 Ecology L.Q. 343, 349-50 (1984) (“[T]he new source review requirements are triggered not only when an operator builds a new plant, but also whenever the operator installs or alters a piece of equipment in an existing plant and thereby increases emissions.”) (emphasis supplied).

WEPCO does not dispute that its steam drum and air heater replacements will result in an altered plant. But WEPCO does assert that Congress did not intend for simple equipment replacement to constitute a physical change for purposes of the Clean Air Act’s modification provisions:

The plain meaning of “modify” is “to change or alter” [Webster’s New World Dictionary] or “to make basic or fundamental changes in.” [Webster’s Ninth New Collegiate Dictionary] Reflecting the plain meaning of this term, Congress provided that a facility (1) must undergo a physical or operational “change” before it is evaluated under the modification provision.... Thus, under the plain meaning of the Act, a unit should not be deemed “modified” as a result of replacement of equipment with equipment similar to that replaced. As in the case of Port Washington, such like-kind replacement does not “change or alter” the design or nature of the facility. Rather, it merely allows the facility to operate again as it had before the specific equipment deteriorated.

Petitioner’s Brief at 32-33.

Chevron instructs us to rely more on congressional direction and on agency construction (pursuant to congressional delegation) than on glosses found in the dictionary. What WEPCO calls “plain” is anything but plain and takes the definition far beyond the words enacted by Congress. Chevron, 467 U.S. at 843-45, 104 S.Ct. at 2781-83; see generally, R. Anthony, Report to the Administrative Conference of the United States: Which Agency Interpretations Should Bind the Courts and the Public? (1989) (explaining Chevron approach). Thus, whether the replacement of air heaters and steam drums is a “basic or fundamental change” in the Port Washington plant is irrelevant for our purposes, given Congress’s directions on the subject: “The term ‘modification’ means any physical change — ” 42 U.S.C. § 7411(a)(4) (emphasis supplied). We follow Congress’s definition of “modification” — not Webster’s —when interpreting this term within the context of the Clean Air Act. Cf. Chevron, 467 U.S. at 861, 104 S.Ct. at 2790 (“[T]he meaning of a word must be ascertained in the. context of achieving particular objectives _”).

Nor can we find any support in the relevant case law for the narrow constructions of “modification” and “physical change” offered by WEPCO. The Supreme Court reported in Chevron that Senator Muskie, one of the principal supporters of the Clean Air Act, remarked: “A source ... is subject to all the nonattainment requirements as a modified source if it makes any physical change which increases the amount of any air pollutant_” 467 U.S. at 853, 104 S.Ct. at 2787 (quoting 123 Cong.Rec. 26847 (1977)) (emphasis supplied). And other courts considering the modification provisions of NSPS and PSD have assumed that “any physical change” means precisely that. See, e.g., National-Southwire Aluminum Co. v. EPA, 838 F.2d 835 (6th Cir.), cert. denied, - U.S.-, 109 S.Ct. 390, 102 L.Ed.2d 379 (1988) (turning off pollution control equipment constitutes “physical change” and modification); Alabama Power Co. v. Costle, 636 F.2d 323, 400 (D.C.Cir.1979) (“[T]he term ‘modification’ is nowhere limited to physical changes exceeding a certain magnitude.”); ASARCO Inc. v. EPA, 578 F.2d 319, 322 (D.C.Cir. 1978) (NSPS applies to any stationary source that is “physically or operationally changed in such a way that its emission of any air pollutant increases.”) (emphasis removed). Cf. United States v. Narragansett Improvement Co., 571 F.Supp. 688, *909694-95 (D.R.I.1983) (replacement program not modification because, despite physical change, no increase in emissions).

Further, to adopt WEPCO’s definition of “physical change” would open vistas of indefinite immunity from the provisions of NSPS and PSD. Were we to hold that the replacement of major generating station systems — including steam drums and air heaters — does not constitute a physical change (and is therefore not a modification), the application of NSPS and PSD to important facilities might be postponed into the indefinite future. There is no reason to believe that such a result was intended by Congress. The Clean Air Act Amendments were enacted to “speed up, expand, and intensify the war against air pollution in the United States with a view to assuring that the air we breathe throughout the Nation is wholesome once again.” H.R. Rep. No. 91-1146, 91st Cong., 2d Sess. 1, 1, reprinted in 1970 U.S.Code Cong. & Admin.News 5356, 5356. In particular, the permit program established by the 1977 Amendments to the Clean Air Act represented a balance between “the economic interests in permitting capital improvements to continue and the environmental interest in improving air quality.” Chevron, 467 U.S. at 851, 104 S.Ct. at 2786. The House echoed this theme in its- Committee report: “[The compliance program is designed, in part,] to allow reasonable economic growth to continue in an area while making reasonable further progress to assure attainment of the [pollution-control] standards by a fixed date_” H.R.Rep. No. 294, 95th Cong., 1st Sess. 211, reprinted in 1977 U.S.Code Cong. & Admin.News 1077, 1290. A too restrictive interpretation of “modification” might upset the economic-environmental balance in unintended ways.

Consistent with its balanced approach, Congress chose not to subject existing plants to the requirements of NSPS and PSD. Members of the House recognized that “[b]uilding control technology into new plants at time of construction will plainly be less costly then [sic] requiring retrofit when pollution control ceilings are reached.” H.R.Rep. No. 294, 95th Cong., 1st Sess. 185, reprinted in 1977 U.S.Code Cong. & Admin.News at 1264. But Congress did not permanently exempt existing plants from these requirements; section 7411(a)(2) provides that existing plants that have been modified are subject to the Clean Air Act programs at issue here. As Judge Boggs, dissenting in National-Southwire, reasoned: “The purpose of the ‘modification’ rule is to ensure that pollution control measures are undertaken when they can be most effective, at the time of new or modified construction. See 116 Cong.Rec. 32,-918. (remarks of Sen. Cooper), reprinted in 1 Senate Committee on Public Works, A Legislative History of the Clean Air Act Amendments of 1970 (1974), at 260.” National-Southwire Aluminum Co. v. EPA, 838 F.2d 835, 843 (6th Cir.) (Boggs, J., dissenting), cert. denied, - U.S.-, 109 S.Ct. 390, 102 L.Ed.2d 379 (1988). Judge Boggs argued that the shutting down of pollution control equipment in an existing plant should not be considered a modification because it would not afford the utility an opportunity for “effective placement of new control technology.” Id. Here the record is silent on this point (although the point is important). How easy or difficult would be “the effective placement of new control technology” in these renovated units is not clear, but we do know that the project already contemplates replacement of steam drums, air heaters and other components; each unit would, therefore, in any event be shut down for nine months.

Our reading of the phrase “any physical change” is also consistent with another of the basic goals of the 1977 Amendments: technology-forcing. The legislative history suggests and courts have recognized that in passing the Clean Air Act Amendments, Congress intended to stimulate the advancement of pollution control technology. See, e.g., S.Rep. No. 91-1196, 91st Cong., 2d Sess. 17 (1970) (“Standards of performance should provide an incentive for industries to work toward constant improvement in techniques for preventing and controlling emissions from stationary sources.... ”); Duquesne Light Co. v. EPA, 698 F.2d 456, 475 (D.C.Cir.1983); *910Alabama Power, 636 F.2d at 372; ASARCO, 578 F.2d at 327; United States v. SCM Corp., 667 F.Supp. 1110, 1126-27 (D.Md. 1987). The development of emissions control systems is not furthered if operators could, without exposure to the standards of the 1977 Amendments, increase production (and pollution) through the extensive replacement of deteriorated generating systems.

2. Increase in Emissions

The controversy involving WEPCO’s alleged increase in emissions primarily concerns the regulations, not the statute: WEPCO argues that the EPA’s regulatory method of measuring emissions is arbitrary and capricious. From a statutory standpoint, however, the modification provisions of the Clean Air Act Amendments are activated once a physical change is coupled with an “increase[ ] [in] the amount of any air pollutant emitted.’’ 42 U.S.C. § 7411(a)(4). See, e.g., United States v. Narragansett Improvement Co., 571 F.Supp. at 694. In the case before us, WEPCO does not dispute that its replacement program — intended to enable its deteriorated generators to operate at full capacity — will cause its emissions to increase from their current operating levels. The question for resolution, however, is whether the EPA properly construed its regulations by comparing actual emission rates with so-called “baseline” rates to determine the increase in emissions for NSPS and PSD purposes. We will discuss this subject later; but for purposes of the statutory requirement, we simply observe that the rejuvenated Port Washington plant will produce more emissions after the completion of the renovation project than the operating deteriorated plant produced shortly before the project was undertaken.

B. The EPA’s Regulations

Although we have determined that WEP-CO’s repair and replacement program satisfies the modification provisions of the Clean Air Act Amendments, this is not the end of our inquiry. WEPCO’s attack focuses primarily on EPA regulations, which in a number of respects are narrower than the statute. WEPCO argues that the EPA applied its regulations arbitrarily and capriciously to the Port Washington project.

1. Physical Change and the “Routine” Exception

EPA regulations define “modification” as “any physical or operational change to an existing facility which results in an increase in the emission rate to the atmosphere of any pollutant to which a standard applies.” 40 C.F.R. § 60.14(a) (1988). To a major degree, this definition parallels 42 U.S.C. section 7411(a)(2), and it is unnecessary to repeat the analysis already applied to the statute. See supra 111(A)(1). However, the EPA has, in addition, used its regulations to exempt a number of activities from the broader definition. The exemption that may be relevant here is accomplished by the following language:

The following shall not, by themselves, be considered modifications under this part:

(1) Maintenance, repair, and replacement which the Administrator determines to be routine for a source category....

40 C.F.R. § 60.14(e) (1988). See 40 C.F.R. § 52.21(b)(2)(iii). WEPCO relies on this language to argue that, even if its repair and replacement program amounts to a physical change, it was specifically exempted by the regulations.

Again, we accord substantial deference to an agency’s interpretation of its own regulations, especially with respect to technical and complex matters. Lyng v. Payne, 476 U.S. 926, 939, 106 S.Ct. 2333, 2341, 90 L.Ed.2d 921 (1986); Aluminum Co. of Am. v. Central Lincoln Peoples’ Util. Dist., 467 U.S. 380, 390, 104 S.Ct. 2472, 2479, 81 L.Ed.2d 301 (1984). In this connection, to determine whether proposed work at a facility is routine, “EPA makes a case-by-case determination by weighing the nature, extent, purpose, frequency, and cost of the work, as well as other relevant factors, to arrive at a common-sense finding.” Clay Memorandum at 3. The EPA considered all these factors in determining that the Port Washington project was not *911routine; first, the EPA observed that the nature and extent of the project was substantial: WEPCO proposed to replace sixty-foot steam drums (in units 2, 3, 4 and 5) and air heaters (in units 1, 2, 3 and 4) during successive nine-month outages at each unit. Id. at 4. Certainly, the magnitude of the project (as well as the downtime required to implement it) suggests that it is more than routine.

Further, the EPA points to WEPCO’s admission in its application that “[work items] falling into the category of repetitive maintenance that are normally performed during scheduled equipment outages ... are not included in this application.” Cassidy Letter at 1 (emphasis supplied). This admission suggests that WEP-CO at first blush did not regard the repair and replacement project as ordinary or routine.

In addition, the EPA noted that far from being routine, the Port Washington project apparently was unprecedented: “WEPCO did not identify, and EPA did not find, even a single instance of renovation work at any electric utility generating station that approached the Port Washington life extension project in nature, scope or extent.” Respondent’s Brief at 44; see Clay Memorandum at 4 (“[T]his is a highly unusual, if not unprecedented, and costly project.”). We surmise, although the record is silent, that the “case of first impression” character of the project may reflect historical practice in the electric utility industry of replacing old plants (at the expiration of their useful lives) with new plants, employing improved technologies and achieving improved efficiencies. This was the typical practice, rather than the mere extension of life of existing plants through massive like-kind replacements. Cf Clay Memorandum at 4 (“[The Port Washington project’s] purpose is to completely rehabilitate aging power generating units whose capacity has significantly deteriorated over a period of years, thereby restoring their original capacity and substantially extending the period of their utilization as an alternative to retiring them as they approach the end of their useful physical and economic life.”).

WEPCO asks us to overlook the factors outlined in the Clay Memorandum and reverse the EPA primarily on the basis of earlier EPA decisions characterizing certain replacement programs as routine; WEPCO argues that the nature and extent of these "routine" projects parallel those of its Port Washington project. For example, WEPCO presented the EPA with a list of forty air heaters in other plants that had been replaced without triggering NSPS or PSD provisions. Letter from Mark P. Steinberg, Superintendent-Air Quality of WEPCO, to Dale Ziege, Wisconsin Department of Natural Resources (Jan. 11, 1989). But as WEPCO has acknowledged, the plate-type air heaters at issue in the Port Washington project must be replaced in whole; in contrast, the forty units where replacement was apparently considered routine contained a Ljungstrom basket or tubular type heater. The Ljungstrom basket type, at least, permits the replacement of the heat transfer surface without requiring the removal of the entire unit. Supplemental Determination at 6-7. Obviously, the precise nature of the physical change is a material factor in determining whether the change is routine, and for this purpose it is important that the subject of past EPA practice be closely comparable with the change under consideration here. See Thomas Letter at 3 ("PSD and NSPS applicability determinations are made on a case-by-case basis."). WEPCO has not demonstrated that the EPA's conclusion that the forty other air heater replacements were dissimilar is arbitrary and capricious.2

The purpose, frequency and cost of the work also support the EPA’s decision here.

*912WEPCO admits that the plans for extensive renovation “represent a life extension of the units from their planned retirement dates,” Cassidy Letter at 2-3 (emphasis supplied), and it recognizes that “the renovation work items included in this application are those that would normally occur only once or twice during a unit’s expected life cycle.” Id. at 1. Indeed, WEPCO reported that it had never previously replaced a steam drum or “header” of comparable size at any of its coal-fired electrical generating facilities. Clay Memorandum at 5. Further, the Port Washington renovation project will cost at least $70.5 million. Letter from John W. Boston, Senior Vice President of WEPCO, to Gary D. McCutchen, Chief New Source Review Section of the EPA, at 4 (May 19, 1988). These factors suggest that the project is not routine.

WEPCO urges that the EPA’s conclusions are supported by neither the evidence nor the provisions of the Clean Air Act Amendments. WEPCO reasons that because any replacement project will presumably extend the life of a facility, the EPA’s reliance on life extension as a factor in denying the “routine” nature of a project is overbroad. Petitioners’ Brief at 44. Although perhaps persuasive on its face, WEPCO’s analysis is ultimately wide of the mark. While it is certainly true that the repair of deteriorated equipment will contribute to the useful life of any facility, it does not necessarily follow that the repairs in question would extend the life expectancy of the facility. The need for some repairs along the line is a given in determining in the first instance the life expectancy of a plant. WEPCO cannot seriously argue that its units’ planned retirement dates of 1992 (units 1 and 2) and 1999 (units 3, 4 and 5) did not take into account at least minor equipment repairs and replacements.3 And WEPCO concedes that the Port Washington program will extend the life expectancy of the plant until 2010. The EPA concluded that the proposed project will increase the life expectancy of the Port Washington facility, and this conclusion was a factor in the finding that the work was not routine. These determinations were not arbitrary and capricious.

Still, WEPCO asserts that the cost, magnitude and nature of its Port Washington project are irrelevant for purposes of the “routine” exception to NSPS and PSD. WEPCO contends that the EPA has already addressed these factors — including the perpetuation of existing sources— through its so-called “reconstruction” rule:

(a) An existing facility, upon reconstruction, becomes an affected facility [subject to NSPS], irrespective of any change in emission rate.

(b) “Reconstruction” means the replacement of components of an existing facility to such an extent that:

(1) The fixed capital cost of the new components exceeds 50 percent of the fixed capital cost that would be required to construct a comparable entirely new facility....

40 C.F.R. § 60.15 (1988).4 See United States Environmental Protection Agency, Electric Utility Steam Generating Units: Background Information for Proposed Particulate Matter Emission Standards 5-7 (1978). WEPCO believes that, because the air heater replacements will presumably cost less than six percent of a wholly new facility, the reconstruction provisions are not triggered. Therefore, WEPCO argues that the cost and scope of the project are relevant only to a “reconstruction” analysis and are not material for purposes of the routine exception to the modification provisions. Petitioner’s Brief at 46.

*913WEPCO’s analysis fails to note, however, the fundamental differences distinguishing the reconstruction and modification provisions. The reconstruction provision applies to any substantial replacement (more than 50% of the cost of a new facility) even if the replacement causes no subsequent increase in emissions. In sharp contrast, the modification provisions apply only when a physical change is accompanied by an increase in emissions. To argue, therefore, that the reconstruction provision is the exclusive determinant of whether the cost, nature and magnitude of a project will require the application of NSPS is to ignore the substantially different objectives of the reconstruction and modification provisions: The reconstruction provision is aimed principally at “discouragpng] the perpetuation of a facility, instead of replacing it at the end of its useful life with a newly constructed affected facility,” without regard to emissions, 39 Fed.Reg. 36946, 36948 (1974), while the modification provision applies to any physical change, without regard to cost, that causes an increase in emissions. See, e.g., ASARCO Inc. v. EPA, 578 F.2d 319 (D.C.Cir.1978); United States v. Narragansett Improvement Co., 571 F.Supp, 688, 695 (D.R.I.1983) (“a ‘reconstruction’ of an existing facility would occur ‘irrespective of any change in emission rate’ upon the replacement of a ‘substantial portion of the existing facility’s components.’ ”). Hence, we cannot agree that the EPA’s consideration of the cost, magnitude and nature of the Port Washington project, for purposes of the modification provision of the regulations (and its “routine” exception), is somehow “preempted” by the reconstruction provisions of the regulations. The EPA’s examination of these factors, therefore, was not arbitrary or capricious.

2. Increase in Emissions

Thus far, we have not had to address the important differences between the PSD and NSPS programs. At this point, however, the differences become crucial, because each program measures emissions in a fundamentally distinct manner.

a. NSPS Measurements

As previously noted, the EPA’s NSPS program is concerned primarily with increases in emission rates, expressed in kilograms per hour of discharged pollutants. 40 C.F.R. § 60.14 (1988). The EPA compares the hourly emissions of the unit at its current maximum capacity to its potential emissions at maximum capacity after the change. Clay Memorandum at 9; see 40 C.F.R. § 60 App. C (1988) (providing complex formulae for determining emission rate change). In this calculation, the agency disregards the unit’s maximum design capacity; this factor often sheds little light on the unit’s actual current capacity to produce emissions.5

The EPA applied these procedures in examining the generating units at Port Washington. The EPA asked WEPCO to submit figures for the actual operations and emissions of each unit at the Port Washington plant for the years 1978 to 1987; the EPA then relied upon the 1987 figures to calculate the emissions baseline against which post-replacement emissions could be compared. WEPCO, however, challenged the EPA’s acceptance of these preliminary baseline figures, arguing that units 1, 2, 3 and 4 were capable of operating at higher rates of production than those calculated by the EPA based upon the 1987 figures. WEPCO conducted five ten-hour tests at each unit to determine its maximum capacity. Upon reviewing the test results, the EPA agreed that units 2 and 3 could be operated at their design capacities, and it revised the baseline levels for these units. The agency concluded that because there would be no increase in production or emissions, NSPS would not apply to these units following the renovation project. Nonetheless, the EPA refused to alter the baseline levels for units 1 and 4, noting that *914WEPCO’s tests had not been conducted pursuant to the test protocol as required by the regulations and the Wisconsin State Implementation Plan (units 1 and 4 exceeded certain maximum allowable emission limits). Supplemental Determination at 8-9. Comparing these 1987 baseline levels to the maximum capacity of the plant after renovation, the EPA concluded that the renovation project would be subject to the provisions of NSPS.

WEPCO asks us to overturn the EPA’s final ruling that the Port Washington project triggers NSPS. Specifically, WEP-CO argues that, by using 1987 figures in determining the emissions baseline, the EPA failed, to apply its own regulations: WEPCO asserts that these figures “reflected voluntary decisions by WEPC[0] regarding safety considerations (e.g., the ‘zero’ rate for Unit 5) and an electricity demand which did not require operation of the units at higher capacities.” Petitioner’s Brief at 15-16. WEPCO also posits that the EPA’s refusal to compare representative pre-renovation emissions with actual post-renovation emissions is contrary to EPA regulations and amounts to an abuse of agency discretion.6

WEPCO’s first assertion is easily dismissed. The EPA’s choice of the 1987 figures was based entirely upon WEPCO’s own data. And, when WEPCO complained that its own data did not reflect WEPCO’s pre-renovation capabilities, the EPA permitted WEPCO to conduct new tests (pursuant to 40 C.F.R. § 60 App. C (1988)) that eventually resulted in the revision of the baselines for units 2 and 3.

WEPCO’s second charge is far more substantial. WEPCO argues that NSPS regulations require the EPA to use a “representative” year in determining a baseline rate of emissions. The EPA disputes this claim, arguing that “[a]s to NSPS, there is no ‘representative emissions’ concept_ Rather ... the baseline emission rates for units 1-5 are determined by hourly maximum capacity just prior to the renovations.”7 Thomas Letter at 5.

WEPCO’s interpretation of the regulations, at first blush, seems sensible: since the regulations require that the manual *915emission tests and continuous monitoring systems be based upon the “representative performance” of the facility, the emission factor test approach must also be based upon “representative performance.” 40 C.F.R. § 60.14 (1988); see 39 Fed.Reg. 36946, 36947 (1974) (explaining provision).8 Otherwise, the tests might reach inconsistent results, making the rate of emissions entirely dependent upon the type of test used by the facility. Hence, argues WEP-CO, the EPA must examine the emission rates during a representative period, not 1987.

WEPCO’s analysis, however, relies upon a flawed premise. WEPCO assumes that the phrase “representative performance of the facility” suggests that the EPA must choose a representative year. Read in context, however, the phrase refers generally to all the conditions of the test, not specifically to its timing:

Tests shall be conducted under such conditions as the Administrator shall specify to the owner or operator based on representative performance of the facility. At least three valid test runs must be conducted before and at least three after the physical or operational change. All operating parameters which may affect emissions must be held constant to the maximum feasible degree for all test runs.

40 C.F.R. § 60.14(b)(2) (1988). Compare 40 C.F.R. § 52.21(b)(21)(ii) (1988) (PSD program) (“The Administrator shall allow the use of a different time period upon a determination that it is more representative of normal source operation.”) (emphasis supplied). Put simply, section 60.14 ensures that the operator will not doctor testing conditions to produce favorable emission results. The EPA’s explanation of its regulations, which of course is given deference, supports this interpretation: “According to the proposed regulation, each set of emission tests (using manual tests or continuous monitors) conducted before and after a physical or operational change would consist of at least three runs, and would be conducted under representative operating conditions.” 39 Fed.Reg. 36946, 36947 (1974) (emphasis supplied). WEPCO has not argued that it conducted its own tests under unrepresentative conditions, nor has it challenged any other part of the test protocol.9 And WEPCO does not claim that the tests were conducted during a period of operations that substantially differed from the normal operations of the deteriorated Port Washington plant. Further, the fact that the EPA permitted WEPCO to conduct additional emissions tests on the units (during which, presumably, WEPCO could maintain representative operating conditions) undermines WEPCO’s assertion that the regulations were applied arbitrarily or capriciously.

b. PSD Measurements

Unlike NSPS, PSD is concerned with changes in total annual emissions, expressed in tons per year. The PSD regulations require preconstruction review of the construction or modification of major emitting facilities. These regulations define their key term — “major modification” — as “any physical change in or change in the method of operation of a major stationary source that would result in a significant10 net emissions increase of any pollutant subject to regulation under the Act.” 40 *916C.F.R. § 52.21(b)(2)(i) (1988) (footnote supplied).

Here the question is whether WEPCO’s renovation project will result in “a significant net emissions increase” so as to trigger the “major modification” provision of the regulations and, as a result, PSD. To determine whether the project would result in an emissions increase, the EPA compared actual pre-renovation emissions with potential post-renovation emissions at the Port Washington plant. Specifically, the EPA first examined the two-year period of 1983 through 1984 as the pre-renovation baseline period, pursuant to 40 C.F.R. section 52.21(b)(21)(ii):

In general, actual emissions as of a particular date shall equal the average rate, in tons per year, at which the unit actually emitted the pollutant during a two-year period which precedes the particular date and which is representative of normal source operation. The Administrator shall allow the use of a different time period upon a determination that it is more representative of normal source operation. Actual emissions shall be calculated using the unit’s actual operating hours, production rates, and types of materials processed, stored, or combusted during the selected time period.

40 C.F.R. § 52.21(b)(21)(ii) (1988) (emphasis supplied). Because Administrator Thomas determined that the discovery of cracks in the rear steam drums led to a more recent “source curtailment,” he relied upon the data from earlier years, 1983 and 1984, as the baseline to determine whether the renovation would cause an increase in emissions. Thomas Letter at 5. WEPCO does not challenge this component of the EPA’s calculation.

Second, the EPA calculated the actual emissions of the plant following completion of the project. Generally, in order to apply PSD, the regulations require the EPA to find an “increase in actual emissions from a particular physical change or change in method of operation.” 40 C.F.R. § 52.21(b)(3)(i)(a) (1988) (emphasis supplied). The EPA reasoned, however, that because the source “ha[d] not yet begun operations following the renovation, ‘actual emissions’ following the renovation [were] deemed to be the source’s ‘potential to emit.’ ” Clay Memorandum at 7. In support of its reliance on WEPCO’s potential to emit, the EPA pointed to the regulations: “For any emissions unit which has not begun normal operations on the particular date, actual emissions shall equal the potential to emit of the unit on that date.” 40 C.F.R. § 52.21(b)(21)(iv) (1988).

WEPCO objects strenuously, and with good reason. In calculating the plant’s post-renovation potential to emit, the EPA bases its figures on round-the-clock operations (24 hours per day, 365 days per year) because WEPCO could potentially operate its facility continuously, despite the fact that WEPCO has never done so in the past. And the EPA has admitted that it “assumed that emissions increases at Port Washington would come not from an increase in emission rate, but rather from increases in production rate or hours of operation.” Supplemental Determination at 9; see Clay Memorandum at 7-8. The EPA responds that WEPCO can avoid these maximum calculations simply by consenting to federally enforceable restrictions on production rates and hours of operation, but WEPCO declines to do so. Clay Memorandum at 8; see 40 C.F.R. §§ 52.21(b)(4), 52.21(b)(16) (1988). Thus, argues the EPA, it has no choice but to assume that the plant will be operated continuously.11

The first issue to be addressed is whether the EPA properly invoked the “potential to emit” concept in calculating the emissions increase. As explained above, *917the PSD regulations state that the EPA may rely upon a facility’s potential to emit if the unit “has not begun normal operations on the particular date.” 40 C.F.R. § 52.21(b)(21)(iv) (1988) (emphasis supplied). WEPCO argues that this phrase should be interpreted to include only those units that have never been in operation, while the EPA urges that the phrase can be applied to both new and modified units.

The regulatory history of this phrase sheds little light on its proper interpretation. The EPA argues that it has always interpreted this phrase to include modified units; it asserts that its formulae for determining emissions increases have consistently assumed that “new or modified units ” would be deemed to operate at maximum physical or federally enforceable levels. 45 Fed.Reg. 52676, 52718 (1980) (emphasis supplied). But the EPA’s analysis here seems circular: in order to demonstrate that the Port Washington like-kind replacement project constitutes a modification, the EPA applies the potential to emit concept (to show an increase in emissions). And in order to apply the potential to emit concept to like-kind replacement, the EPA assumes that the plant is a “modified” unit. Although we accord great deference to an agency construing the statute it administers, Chevron, 467 U.S. at 844, 104 S.Ct. at 2782, and even more deference to an agency interpreting its own complex regulations, Aluminum Co. of Am. v. Central Lincoln Peoples’ Util. Dist., 467 U.S. at 390, 104 S.Ct. at 2479, we cannot defer to agency interpretations that, as applied here, appear to assume what they seek to prove.12

We are also troubled by the EPA’s assumption of continuous operations in calculating potential to emit at the Port Washington plant. Although we agree that the EPA cannot reasonably rely on a utility’s own unenforceable estimates of its annual emissions,13 we find no support in the regulations for the EPA’s decision wholly to disregard past operating conditions at the plant. Indeed, Alabama Power Co. v. Costle, 636 F.2d 323 (D.C.Cir. 1979), which contributes importantly to the EPA’s current PSD program, suggests otherwise. There, the D.C. Circuit held, in part, that the EPA must “take[] into account the anticipated functioning of the air pollution control equipment designed into the facility” when calculating the facility’s potential to emit. Id. at 353. More important for our purposes, however, was the court’s discussion of a unit’s potential to emit:

If the source has no actual emissions because it has yet to commence operating, its hypothetical, projected emissions are included in the baseline. If, however, the source is an established operation, a more realistic assessment of its impact on ambient air quality levels is possible, and thus is directed.

*918Id. at 379 (emphasis supplied). The district court in United States v. Louisiana-Pacific Corp., 682 F.Supp. 1141 (D.Colo.1988), relying on Alabama Power, recently reached the same conclusion:

The broad holding of Alabama Power is that potential to emit does not refer to the maximum emissions that can be generated by a source hypothesizing the worst conceivable operation. Rather, the concept contemplates the maximum emissions that can be generated while operating the source as it is intended to be operated and as it is normally operated. Of course, it is possible that a source could be operated without the control equipment designed into it or that a Ko-nus heater could be operated so badly that the fire would go out. Yet, Alabama Power stands for the proposition that hypothesizing the worst possible emissions from the worst possible operation is the wrong way to calculate potential to emit.

Id. at 1158.

In sum, we certainly do not suggest that the EPA may never subject replaced units to the potential to emit concept under its regulations. The EPA may, if it wishes, undertake notice and comment procedures to apply the potential to emit concept to like-kind replacement. See 42 U.S.C. § 7607(d). But existing regulations do not seem to us to support such an application. We therefore believe that the EPA’s reliance on an assumed continuous operation as a basis for finding an emissions increase is not properly supported. The EPA’s determination that there has been a major modification for PSD purposes must be set aside.14

IV. Fuel Switching

The final significant dispute in this case involves fuel switching. WEPCO proposed to the EPA that its “replacement project combined with an enforceable fuel switch would not ‘result[ ] in an increase in the [sulfur dioxide and particulate matter] emission rate[s]’ from those units.” Petitioner’s Brief at 50 (brackets in original); see 40 C.F.R. § 60.14(a) (1988). Nonetheless, the EPA refused to permit WEP-CO to utilize lower sulfur coal instead of implementing pollution control technologies to prevent an increase in emissions. The EPA explained that “the statute reflects a basic political decision that fossil fuel-fired sources not rely only on natural occurring less-polluting fuels to comply with the NSPS. Instead, Congress declared that compliance must depend in part upon the application of flue gas treatment or other pollution control technologies.” Supplemental Determination at 10. Further, the EPA pointed to 40 C.F.R. section 60.14(b)(2) (1988), which requires that “operating parameters” — including fuel and raw materials — must be held constant in measuring emissions before and after renovations to determine whether the utility has undertaken a modification. WEPCO disputes the EPA’s interpretation of the relevant provisions of the Clean Air Act Amendments.

Consistent with the Supreme Court’s approach in Chevron, we. first examine whether the statute evinces a clear congressional intent on the matter. We believe it does. Although the plain language of 42 U.S.C. section 7411 does not resolve the issue, the relevant legislative history provides ample support for the EPA’s position. The House Conference Report, for example, states:

The agreement requires (1) that the standards of performance for fossil fuel-fired boilers be substantially upgraded to require the use of the best technological system of continuous emission reduction and to preclude use of untreated low sulfur coal alone as a means of compliance; ... (3) that for fossil fuel-fired sources, the new source performance *919standards must be comprised of both a standard of performance for emissions and an enforceable requirement for a percentage reduction in pollution from untreated fuel.

H.R.Rep. No. 564, 95th Cong., 1st Sess. 130, reprinted in 1977 U.S.Code Cong. & Admin.News 1077, 1510 (emphasis supplied). In addition, passages from the congressional debates reflect Congress’s refusal to allow stationary sources to substitute low sulfur fuels to avoid a requirement of pollution control technology. See, e.g., Ill Senate Committee on Environment & Public Works, A Legislative History of the Clean Air Act Amendments of 1977, at 323, 353 (1978) (disapproving substitution of low sulfur coal for pollution control technology); IV Senate Committee on Environment & Public Works, supra, at 2653 (same). In these reports, Congress reasoned that the Administrator’s previous standards — which had allowed fuel switching in lieu of pollution control technology— directly conflicted with the purposes of the NSPS program:

1. The standards give a competitive advantage to those States with cheaper low-sulfur coal and create a disadvantage for Midwestern and Eastern States where predominantly higher sulfur coals are available;

2. These standards do not provide for maximum practicable emission reduction using locally available fuels, and therefore do not maximize potential for long-term growth;

3. These standards do not help to expand the energy resources (that is, higher sulfur coal) that could be burned in compliance with emission limits as intended;

4. These standards aggravate compliance problems for existing coal-burning stationary sources which cannot retrofit and which must compete with larger, new sources for low-sulfur coal;

5. These standards increase the risk of early plant shutdowns by existing plants (for the reasons stated above), with greater risk of unemployment; and

6. These standards operate as a disincentive to the improvement of technology of new sources, since untreated fuels could be burned instead of using such new, more effective technology.

Ill Senate Committee on Environment & Public Works, supra, at 323. These purposes, reflecting technological and political choices, demonstrate that Congress rejected fuel switching as a method of avoiding the impact of NSPS. We believe Congress left us no choice on this issue.

V. CONCLUSION

In an era of increasing environmental concern, Congress enacted the Clean Air Act to “speed up, expand, and intensify the war against air pollution in the United States with a view to assuring that the air we breathe throughout the Nation is wholesome once again.” H.R.Rep. No. 1146, 91st Cong., 2d Sess. 1, 1, 1970 U.S.Code Cong. & Admin.News 5356, 5356. The EPA is entitled to substantial deference in interpreting the technical provisions of the Act and its own regulations. We cannot grant deference, however, where the EPA has attempted to implement the Act’s lofty goals in contravention of its own statutory regime. We therefore affirm in part and vacate in part, remanding the cause to the EPA for further proceedings not inconsistent with this opinion.

*565Justice Souter

delivered the opinion of the Court.

In the 1970s, Congress added two air pollution control schemes to the Clean Air Act: New Source Performance *566Standards (NSPS) and Prevention of Significant Deterioration (PSD), each of them covering modified, as well as new, stationary sources of air pollution. The NSPS provisions define the term “modification,” 42 U. S. C. § 7411(a)(4), while the PSD provisions use that word “as defined in” NSPS, § 7479(2)(C). The Court of Appeals concluded that the statute requires the Environmental Protection Agency (EPA) to conform its PSD regulations on “modification” to their NSPS counterparts, and that EPA’s 1980 PSD regulations can be given this conforming construction. We hold that the Court of Appeals’s reading of the 1980 PSD regulations, intended to align them with NSPS, was inconsistent with their terms and effectively invalidated them; any such result must be shown to comport with the Act’s restrictions on judicial review of EPA regulations for validity.

I

The Clean Air Amendments of 1970, 84 Stat. 1676, broadened federal authority to combat air pollution, see Chevron U. S. A. Inc. v. Natural Resources Defense Council, Inc., 467 U. S. 837, 845-846 (1984), and directed EPA to devise National Ambient Air Quality Standards (NAAQS) limiting various pollutants, which the States were obliged to implement and enforce, 42 U. S. C. §§ 7409, 7410. The amendments dealing with NSPS authorized EPA to require operators of stationary sources of air pollutants to use the best technology for limiting pollution, Chevron, supra, at 846; see also 1 F. Grad, Environmental Law §2.03 [14], p. 2-356 *567(2006), both in newly constructed sources and those undergoing “modification,” 42 U. S. C. § 7411(a)(2). Section 111(a) of the 1970 amendments defined this term within the NSPS scheme as “any physical change in, or change in the method of operation of, a stationary source which increases the amount of any air pollutant emitted by such source or which results in the emission of any air pollutant not previously emitted,” 42 U. S. C. § 7411(a)(4).

EPA’s 1975 regulations implementing NSPS provided generally that “any physical or operational change to an existing facility which results in an increase in the emission rate to the atmosphere of any pollutant to which a standard applies shall be considered a modification within the meaning of section 111.” 40 CFR §60.14(a) (1976). Especially significant here is the identification of an NSPS “modification” as a change that “increase^] . . . the emission rate,” which “shall be expressed as kg/hr of any pollutant discharged into the atmosphere. ” § 60.14(b).1

NSPS, however, did too little to “achiev[e] the ambitious goals of the 1970 Amendments,” R. Belden, Clean Air Act 7 (2001) (hereinafter Belden), and the Clean Air Act Amendments of 1977, 91 Stat. 685, included the PSD provisions, which aimed at giving added protection to air quality in certain parts of the country “notwithstanding attainment and *568maintenance of” the NAAQS. 42 U. S. C. § 7470(1).2 The 1977 amendments required a PSD permit before a “major emitting facility” could be “constructed” in an area covered by the scheme. § 7475(a). As originally enacted, PSD applied only to newly constructed sources, but soon a technical amendment added the following subparagraph: “The term ‘construction’ when used in connection with any source or facility, includes the modification (as defined in section 111(a)) of any source or facility.” §14(a)(54), 91 Stat. 1402, 42 U. S. C. § 7479(2)(C); see also New York v. EPA, 413 F. 3d 3, 13 (CADC 2005) (per curiam). In other words, the “construction” requiring a PSD permit under the statute was made to include (though it was not limited to) a “modification” as defined in the statutory NSPS provisions.

In 1980, EPA issued PSD regulations,3 which “limited the application of [PSD] review” of modified sources to instances of “‘major’ modification,” Belden 46, defined as “any physical change in or change in the method of operation of a major stationary source that would result in a significant net emissions increase of any pollutant subject to regulation under the Act.” 40 CFR § 51.166(b)(2)(i) (1987). Further regulations in turn addressed various elements of this definition, three of which are to the point here. First, the regulations specified that an operational change consisting merely of “[a]n increase in the hours of operation or in the production rate” would not generally constitute a “physical change in or change in the method of operation.” § 51.166(b)(2)(iii)(/). For purposes of a PSD permit, that is, such an operational *569change would not amount to a “modification” as the Act defines it. Second, the PSD regulations defined a “net emissions increase” as “[a]ny increase in actual emissions from a particular physical change or change in the method of operation,” net of other contemporaneous “increases and decreases in actual emissions at the source.” §51.166(b)(3)(i). “Actual emissions” were defined to “equal the average rate, in tons per year, at which the unit actually emitted the pollutant during a two-year period which precedes the particular date and which is representative of normal source operation.” §51.166(b)(21)(ii). “[AJetual emissions” were to be “calculated using the unit’s actual operating hours [and] production rates.” Ibid. Third, the term “significant” was defined as “a rate of emissions that would equal or exceed” one or another enumerated threshold, each expressed in “tons per year.” § 51.166(b)(23)(i).

It would be bold to try to synthesize these statutory and regulatory provisions in a concise paragraph, but three points are relatively clear about the regime that covers this case:

(a) The Act defines modification of a stationary source of a pollutant as a physical change to it, or a change in the method of its operation, that increases the amount of a pollutant discharged or emits a new one.

(b) EPA’s NSPS regulations require a source to use the best available pollution-limiting technology only when a modification would increase the rate of discharge of pollutants measured in kilograms per hour.

(c) EPA’s 1980 PSD regulations require a permit for a modification (with the same statutory definition) only when it is a major one and only when it would increase the actual annual emission of a pollutant above the actual average for the two prior years.

The Court of Appeals held that Congress’s provision defining a PSD modification by reference to an NSPS modifica*570tion caught not only the statutory NSPS definition, but also whatever regulatory gloss EPA puts on that definition at any given time (for the purposes of the best technology requirement). When, therefore, EPA’s PSD regulations specify the “change” that amounts to a “major modification” requiring a PSD permit, they must measure an increase in “the amount of any air pollutant emitted,” 42 U. S. C. § 7411(a)(4), in terms of the hourly rate of discharge, just the way NSPS regulations do. Petitioners and the United States say, on the contrary, that when EPA addresses the object of the PSD scheme it is free to put a different regulatory interpretation on the common statutory core of “modification,” by measuring increased emission not in terms of hourly rate but by the actual, annual discharge of a pollutant that will follow the modification, regardless of rate per hour. This disagreement is the nub of the case.

II

Respondent Duke Energy Corporation runs 30 coal-fired electric generating units at eight plants in North and South Carolina. United States v. Duke Energy Corp., 411 F. 3d 539, 544 (CA4 2005). The units were placed in service between 1940 and 1975, and each includes a boiler containing thousands of steel tubes arranged in sets. Ibid. Between 1988 and 2000,4 Duke replaced or redesigned 29 tube assemblies in order to extend the life of the units and allow them to run longer each day. Ibid.

*571The United States filed this action in 2000, claiming, among other things, that Duke violated the PSD provisions by doing this work without permits. Environmental Defense, North Carolina Sierra Club, and North Carolina Public Interest Research Group Citizen Lobby/Education Fund intervened as plaintiffs and filed a complaint charging similar violations.

Duke moved for summary judgment, one of its positions being that none of the projects was a “major modification” requiring a PSD permit because none increased hourly rates of emissions. The District Court agreed with Duke’s reading of the 1980 PSD regulations. It reasoned that their express exclusion of “ ‘[a]n increase in the hours of operation’ ” from the definition of a “ ‘physical change or change in the method of operation’” implied that “post-project emissions levels must be calculated assuming” preproject hours of operation. 278 F. Supp. 2d 619, 640-641 (MDNC 2003). Consequently, the District Court said, a PSD “major modification” can occur “only if the project increases the hourly rate of emissions.” Id., at 641. The District Court found further support for its construction of the 1980 PSD regulations in one letter and one memorandum written in 1981 by EPA’s Director of the Division of Stationary Source Enforcement, Edward E. Reich. Id., at 641-642.

The United States and intervenor-plaintiffs (collectively, plaintiffs) subsequently stipulated “that they do not contend that the projects at issue in this case caused an increase in the maximum hourly rate of emissions at any of Duke Energy’s units.” App. 504. Rather, their claim “is based solely on their contention that the projects would have been projected to result in an increased utilization of the units at issue.” Ibid. Duke, for its part, stipulated to plaintiffs’ right to appeal the District Court’s determination that projects resulting in greater operating hours are not “major modifications” triggering the PSD permit requirement, absent an increase in the hourly rate of emissions. The Dis*572trict Court then entered summary judgment for Duke on all PSD claims.

The Court of Appeals for the Fourth Circuit affirmed, “albeit for somewhat different reasons.” 411 F. 3d, at 542. “[T]he language and various interpretations of the PSD regulations . . . are largely irrelevant to the proper analysis of this case,” reasoned the Court of Appeals, “because Congress’ decision to create identical statutory definitions of the term ‘modification’” in the NSPS and PSD provisions of the Clean Air Act “has affirmatively mandated that this term be interpreted identically” in the regulations promulgated under those provisions. Id., at 547, n. 3, 550. The Court of Appeals relied principally on the authority of Rowan Cos. v. United States, 452 U. S. 247, 250 (1981), where we held against the Government’s differing interpretations of the word “wages” in different tax provisions. 411 F. 3d, at 550. As the Court of Appeals saw it, Rowan establishes an “effectively irrebuttable” presumption that PSD regulations must contain the same conditions for a “modification” as the NSPS regulations, including an increase in the hourly rate of emissions.5 411 F. 3d, at 550.

As the Court of Appeals said, Duke had not initially relied on Rowan, see 411 F. 3d, at 547, n. 4, and when the Court sua sponte requested supplemental briefing on Rowan's relevance, plaintiffs injected a new issue into the case. They argued that a claim that the 1980 PSD regulation exceeded statutory authority would be an attack on the validity of the regulation that could not be raised in an enforcement proceeding. See 42 U. S. C. § 7607(b)(2). Under § 307(b) of the *573Act, they said, judicial review for validity can be obtained only by a petition to the Court of Appeals for the District of Columbia Circuit, generally within 60 days of EPA’s rule-making. 42 U. S. C. § 7607(b).

The Court of Appeals rejected this argument. “Our choice of this interpretation of the PSD regulations ... is not an invalidation of those regulations,” it said, because “the PSD regulations can be interpreted” to require an increase in the hourly emissions rate as an element of a major “modification” triggering the permit requirement. 411 F. 3d, at 549, n. 7. To show that the 1980 PSD regulations are open to this construction, the Court of Appeals cited the conclusions of the District Court and the Reich opinions.

We granted the petition for certiorari brought by intervenor-plaintiffs, 547 U. S. 1127 (2006), and now vacate.

III

The Court of Appeals understood that it was simply construing EPA’s 1980 PSD regulations in a permissible way that left them in harmony with their NSPS counterpart and, hence, the Act’s single definition of “modification.” The plaintiffs say that the Court of Appeals was rewriting the PSD regulations in a way neither required by the Act nor consistent with their own text.

It is true that no precise line runs between a purposeful but permissible reading of the regulation adopted to bring it into harmony with the Court of Appeals’s view of the statute, and a determination that the regulation as written is invalid. But the latter occurred here, for the Court of Appeals’s efforts to trim the PSD regulations to match their different NSPS counterparts can only be seen as an implicit declaration that the PSD regulations were invalid as written.

A

In applying the 1980 PSD regulations to Duke’s conduct, the Court of Appeals thought that, by defining the term *574“modification” identically in its NSPS and PSD provisions, the Act required EPA to conform its PSD interpretation of that definition to any such interpretation it reasonably adhered to under NSPS. But principles of statutory construction are not so rigid. Although we presume that the same term has the same meaning when it occurs here and there in a single statute, the Court of Appeals mischaracterized that presumption as “effectively irrebuttable.” 411 F. 3d, at 550. We also understand that “[m]ost words have different shades of meaning and consequently may be variously construed, not only when they occur in different statutes, but when used more than once in the same statute or even in the same section.” Atlantic Cleaners & Dyers, Inc. v. United States, 286 U. S. 427, 433 (1932). Thus, the “natural presumption that identical words used in different parts of the same act are intended to have the same meaning ... is not rigid and readily yields whenever there is such variation in the connection in which the words are used as reasonably to warrant the conclusion that they were employed in different parts of the act with different intent.” Ibid. A given term in the same statute may take on distinct characters from association with distinct statutory objects calling for different implementation strategies.

The point is the same even when the terms share a common statutory definition, if it is general enough, as we recognized in Robinson v. Shell Oil Co., 519 U. S. 337 (1997). There the question was whether the term “employees” in § 704(a) of Title VII of the Civil Rights Act of 1964 covered former employees. Title VII expressly defined the term “employee,” 42 U. S. C. § 2000e(f), but the definition was “consistent with either current or past employment,” 519 U. S., at 342, and we held that “each section” of Title VII “must be analyzed to determine whether the context gives the term a further meaning that would- resolve the issue in dispute,” id., at 343-344.

*575If Robinson were inconsistent with Rowan (on which the Court of Appeals relied), it would be significant that Robinson is the later case, but we read the two as compatible. In Rowan, the question was whether the value of meals and lodging given to employees by an employer for its own convenience should be counted in computing “wages” under the Federal Insurance Contributions Act (FICA), 26 U. S. C. § 3101 et seq. (2000 ed. and Supp. IV), and the Federal Unemployment Tax Act (FUTA), 26 U. S. C. § 3301 et seq. (2000 ed. and Supp. IV). Treasury Regulations made this value “includable in ‘wages’ as defined in FICA and FUTA, even though excludable from ‘wages’ under the substantially identical” statutory definition of “wages” for income-tax withholding purposes. 452 U. S., at 252. Although we ultimately held that the income-tax treatment was the proper one across the board, we did not see it this way simply because a “substantially identical” definition of “wages” appeared in each of the different statutory provisions. Instead, we relied on a manifest “congressional concern for the interest of simplicity and ease of administration.” Id., at 255 (internal quotation marks omitted). The FICA and FUTA regulations fell for failing to “serve that interest,” id., at 257, not for defying definitional identity.

In fact, in a setting much like Rowan, we recently declined to require uniformity when resolving ambiguities in identical statutory terms. In United States v. Cleveland Indians Baseball Co., 532 U. S. 200 (2001), we rejected the notion that using the phrase “wages paid” in both “the discrete taxation and benefits eligibility contexts” can, standing alone, “compel symmetrical construction,” id., at 213; we gave “substantial judicial deference” to the “longstanding,” “reasonable,” and differing interpretations adopted by the Internal Revenue Service in its regulations and Revenue Rulings. Id., at 218-220. There is, then, no “effectively irrebuttable” presumption that the same defined term in different provisions of the *576same statute must “be interpreted identically.” 411 F. 3d, at 550. Context counts.

It is true that the Clean Air Act did not merely repeat the term “modification” or the same definition of that word in its NSPS and PSD sections; the PSD language referred back to the section defining “modification” for NSPS purposes. 42 U. S. C. §7479(2)(C).But that did not matter in Robinson, and we do not see the distinction as making any difference here. Nothing in the text or the legislative history of the technical amendments that added the cross-reference to NSPS suggests that Congress had details of regulatory implementation in mind when it imposed PSD requirements on modified sources; the cross-reference alone is certainly no unambiguous congressional code for eliminating the customary agency discretion to resolve questions about a statutory definition by looking to the surroundings of the defined term, where it occurs. See New York, 413 F. 3d, at 19 (“So far as appears,... [this] incorporation by reference [is] the equivalent of Congress’s having simply repeated in the [PSD] context the definitional language used before in the NSPS context”); cf. 91 Stat. 745 (expressly incorporating in an unrelated provision of the 1977 amendments “the interpretative regulation of the [EPA] Administrator . .. published in 41 Federal Register 55524-30” with specified exceptions); New York, supra, at 19 (“Congress’s failure to use such an express incorporation of prior regulations for ‘modification’ cuts against” any suggestion that “Congress intended to incorporate” into the Act the “preexisting regulatory definition” of “modification”). Absent any iron rule to ignore the reasons for regulating PSD and NSPS “modifications” differently, EPA’s construction need do no more than fall within the limits of what is reasonable, as set by the Act’s common 6 definition.

*577B

The Court of Appeals’s reasoning that the PSD regulations must conform to their NSPS counterparts led the court to read those PSD regulations in a way that seems to us too far a stretch for the language used. The 1980 PSD regulations on “modification” simply cannot be taken to track the Agency’s regulatory definition under the NSPS.

True, the 1980 PSD regulations may be no seamless narrative, but they clearly do not define a “major modification” in terms of an increase in the “hourly emissions rate.” On its face, the definition in the PSD regulations specifies no rate at all, hourly or annual, merely requiring a physical or operational change “that would result in a significant net emissions increase of any” regulated pollutant. 40 CFR § 51.166(b)(2)(i). But even when a rate is mentioned, as in the regulatory definitions of the two terms, “significant” and “net emissions increase,” the rate is annual, not hourly. Each of the thresholds that quantify “significant” is described in “tons per year,” § 51.166(b)(23)(i), and a “net emissions increase” is an “increase in actual emissions” measured against an “average” prior emissions rate of so many “tons *578per year,” §§51.166(b)(3)(i) and (21)(ii). And what is further at odds with the idea that hourly rate is relevant is the mandate that “[ajctual emissions shall be calculated using the unit’s actual operating hours,” § 51.166(b)(21)(ii), since “actual emissions” must be measured in a manner that looks to the number of hours the unit is or probably will be actually running. What these provisions are getting at is a measure of actual operations averaged over time, and the regulatory language simply cannot be squared with a regime under which “hourly rate of emissions,” 411 F. 3d, at 550 (emphasis deleted), is dispositive.

The reasons invoked by the Court of Appeals for its different view are no match for these textual differences. The appellate court cited two authorities ostensibly demonstrating that the 1980 PSD regulations “can be interpreted consistently” with the hourly emissions test, the first being the analysis of the District Court in this case. Id., at 549, n. 7. The District Court thought that an increase in the hourly emissions rate was necessarily a prerequisite to a PSD “major modification” because a provision of the 1980 PSD regulations excluded an “ ‘increase in the hours of operation or in the production rate’ ” from the scope of “ ‘[a] physical change or change in the method of operation.’ ” 278 F. Supp. 2d, at 640-641 (quoting 40 CFR §§51.166(b)(2)(iii)(/) and (3)(i)(a) (1987)). The District Court read this exclusion to require, in effect, that a source’s hours of operation “be held constant” when preproject emissions are being compared with postproject emissions for the purpose of calculating the “net emissions increase.” 278 F. Supp. 2d, at 640.

We think this understanding of the 1980 PSD regulations makes the mistake of overlooking the difference between the two separate components of the regulatory definition of “major modification”: “[1] any physical change in or change in the method of operation of a major stationary source that [2] would result in a significant net emissions increase *579of any pollutant subject to regulation under the Act.” § 51.166(b)(2)(i); cf. New York, 413 F. 3d, at 11 (“[The statutory] definition requires both a change — whether physical or operational — and a resulting increase in emissions of a pollutant” (emphasis in original)); Wisconsin Elec. Power Co. v. Reilly, 893 F. 2d 901, 907 (CA7 1990) (same). The exclusion of “increase in . . . hours . . . or . . . production rate,” §51.166(b)(2)(iii)(/), speaks to the first of these components (“physical change ... or change in . . . method,” § 51.166(b)(2)(i)), but not to the second (“significant net emissions increase,” ibid.). As the preamble to the 1980 PSD regulations explains, forcing companies to obtain a PSD permit before they could simply adjust operating hours “would severely and unduly hamper the ability of any company to take advantage of favorable market conditions.” 45 Fed. Reg. 52704. In other words, a mere increase in the hours of operation, standing alone, is not a “physical change or change in the method of operation.” 40 CFR §51.166(b)(2)(iii).

But the District Court took this language a step further. It assumed that increases in operating hours (resulting in emissions increases at the old rate per hour) must be ignored even if caused or enabled by an independent “physical change ... or change in the method of operation.” §51.166(b)(2)(i). That reading, however, turns an exception to the first component of the definition into a mandate to ignore the very facts that would count under the second, which defines “net emissions increase” in terms of “actual emissions,” §51.166(b)(3)(i), during “the unit’s actual operating hours,” § 51.166(b)(21)(ii); see also 57 Fed. Reg. 32328 (1992) (“[A]n increase in emissions attributable to an increase in hours of operation or production rate which is the result of a construction-related activity is not excluded from [PSD] review .. . ”).7

*580The Court of Appeals invoked one other source of support, the suggestion in the Reich opinions that a physical or operational change increasing a source’s hours of operation, without an increase in the hourly emissions rate, cannot be a PSD “major modification.” Duke continues to rely on those opinions here, asserting that “there are no contrary Agency pronouncements.” Brief for Respondent Duke 28. The Reich letters are not, however, heavy ammunition. Their persuasiveness is elusive, neither of them containing more than one brief and conelusory statement supporting Duke’s position. Nor, it seems, are they unembarrassed by any “contrary Agency pronouncements.” See, e.g., App. 258 (Memorandum of Don R. Clay, Acting Assistant EPA Administrator for Air and Radiation (Sept. 9,1988) (when “plans to increase production rate or hours of operation are inextricably intertwined with the physical changes planned,” they are “precisely the type of change in hours or rate o[f] operation that would disturb a prior assessment of a source’s environmental impact and should have to undergo PSD review scrutiny” (internal quotation marks and alterations omitted))); see also 57 Fed. Reg. 32328. In any event, it answers the citation of the Reich letters to realize that an isolated opinion of *581an agency official does not authorize a court to read a regulation inconsistently with its language.8

In sum, the text of the 1980 PSD regulations on “modification” doomed the Court of Appeals’s attempt to equate those regulations with their NSPS counterpart. As a consequence, we have to see the Court of Appeals’s construction of the 1980 PSD regulations as an implicit invalidation of those regulations, a form of judicial review implicating the provisions of § 307(b) of the Act, which limit challenges to the validity of a regulation during enforcement proceedings when such review “could have been obtained” in the Court of Appeals for the District of Columbia Circuit within 60 days of EPA rulemaking. See 42 U. S. C. § 7607(b); see also United States v. Cinergy Corp., 458 F. 3d 705, 707-708 (CA7 2006); Wisconsin Elec. Power Co., 893 F. 2d, at 914, n. 6. Because the Court of Appeals did not believe that its analysis reached validity, it did not consider the applicability or effect of that limitation here. We have no occasion at this point to consider the significance of § 307(b) ourselves.

IV

Finally, Duke assumes for argument that the Act and the 1980 regulations may authorize EPA to construe a PSD “modification” as it has done, but it charges that the agency has taken inconsistent positions and is now “retroactively *582targeting twenty years of accepted practice.” Brief for Respondent Duke 37; see also Brief for State of Alabama et al. as Amici Curiae. This claim, too, has not been tackled by the District Court or the Court of Appeals; to the extent it is not procedurally foreclosed, Duke may press it on remand.

* * *

The judgment of the Court of Appeals is vacated, and the case is remanded for further proceedings consistent with this opinion.

It is so ordered.

Justice Thomas,

concurring in part.

I join all but Part III-A of the Court’s opinion. I write separately to note my disagreement with the dicta in that portion of the opinion, which states that the statutory cross-reference does not mandate a singular regulatory construction.

The Prevention of Significant Deterioration (PSD) statute explicitly links the definition of the term “modification” to that term’s definition in the New Source Performance Standard (NSPS) statute:

“The term ‘construction’ when used in connection with any source or facility, includes the modification (as defined in section 7411(a) of this title) of any source or facility.” 42 U. S. C. § 7479(2)(C).

Section 7411(a) contains the NSPS definition of “modification,” which the parties agree is the relevant statutory definition of the term for both PSD and NSPS. Because of the cross-reference, the definitions of “modification” in PSD and NSPS are one and the same. The term “modification” therefore has the same meaning despite contextual variations in the two admittedly different statutory schemes. Congress’ explicit linkage of PSD’s definition of “modification” to NSPS’ prevents the Environmental Protection *583Agency (EPA) from adopting differing regulatory definitions of “modification” for PSD and NSPS. Cf. IBP, Inc. v. Alvarez, 546 U. S. 21, 34 (2005) (concluding that an “explicit reference” to a previous statutory definition prohibits interpreting the same word differently).

Section 7479(2)(C)’s cross-reference carries more meaning than the mere repetition of the same word in a different statutory context. When Congress repeats the same word in a different statutory context, it is possible that Congress might have intended the context to alter the meaning of the word. See Atlantic Cleaners & Dyers, Inc. v. United States, 286 U. S. 427, 433 (1932). No such possibility exists with §7479(2)(C). By incorporating NSPS’ definition of “modification,” Congress demonstrated that it did not intend for PSD’s definition of “modification” to hinge on contextual factors unique to the PSD statutory scheme. Thus, United States v. Cleveland Indians Baseball Co., 532 U. S. 200 (2001), which analyzes the mere repetition of the same word in a different statutory context, carries little weight in this situation.

Likewise, this case differs from the circumstance we faced in Robinson v. Shell Oil Co., 519 U. S. 337 (1997). In Robinson, we considered whether “employee,” as used in § 704(a) of Title VII of the Civil Rights Act of 1964, included former employees. We determined that under the clear language of the statute, certain statutory provisions using the term “employee” made sense only with respect to former employees or current employees, but not both. Id., at 342-343. Accordingly, upon analyzing the context of § 704(a), we were compelled to conclude that the term “employee” included former employees. This case does not present a similar situation. The statute here includes a statutory cross-reference, which conveys a clear congressional intent to provide a common definition for the term “modification.” And the contextual differences between PSD and NSPS do not compel *584different meanings for the term “modification.” Robinson is, therefore, inapplicable.

Even if the cross-reference were merely the equivalent of repeating the words of the definition, we must still apply our usual presumption that the same words repeated in different parts of the same statute have the same meaning. See Atlantic Cleaners, supra, at 483; ante, at 574. That presumption has not been overcome here. While the broadly stated regulatory goals of PSD and NSPS differ, these contextual differences do not compel different definitions of “modification.” That is, unlike in Robinson, reading the statutory definition in the separate contexts of PSD and NSPS does not require different interpretations of the term “modification.” EPA demonstrated as much when it recently proposed regulations that would unify the regulatory definitions of “modification.” See 70 Fed. Reg. 61083, n. 3 (2005) (terming the proposal “an appropriate exercise of our discretion” and stating that the unified definition better serves PSD’s goals).

The majority opinion does little to overcome the presumption that the same words, when repeated, carry the same meaning. Instead, it explains that this Court’s cases do not compel identical language to be interpreted identically in all situations. Granting that point, the majority still has the burden of stating why our general presumption does not control the outcome here. It has not done so.

Opinion for the court filed by Circuit Judge J. SKELLY WRIGHT.

Concurring opinion filed by Circuit Judge LEVENTHAL.

Circuit Judge MacKINNON filed an opinion concurring in part and dissenting in part on February 28, 1978.

J. SKELLY WRIGHT, Circuit Judge:

These cases involve challenges by ASAR-CO Incorporated, Newmont Mining Corporation, and Magma Copper Company (hereinafter referred to collectively as ASARCO) and the Sierra Club (Sierra)1 to regulations *321issued by the Environmental Protection Agency (EPA).2 The challenged provisions modify previous regulations implementing Section 111 of the Clean Air Act, 42 U.S.C. § 1857c-6 (1970 & Supp. V 1975), as amended, Pub.L.No. 95-95 § 109, 91 Stat. 685, 697-703 (1977), which mandates national emission standards for new stationary sources of air pollution, by introducing a limited form of what the parties call the “bubble concept.”3 This court has jurisdiction over these petitions under Section 307(b)(1) of the Act, 42 U.S.C. § 1857h-5(b)(1) (1970 & Supp. V 1975), as amended, Pub.L.No. 95-95 § 305, 91 Stat. 772-777.

I

A. Section 111 and the “Bubble Concept”

The 1970 amendments to the Clean Air Act4 were passed in reaction to the failure of the states to cooperate with the federal government in effectuating the stated purposes of the Act, especially the commitment “to protect and enhance the quality of the Nation’s air resources so as to promote the public health and welfare and the productive capacity of its population.” Clean Air Act § 101(b)(1), 42 U.S.C. § 1857(b)(1) (1970). See generally W. Rogers, Environmental Law § 3.1 (1977). The 1970 changes were designed “to improve the quality of the nation’s air,” 84 Stat. 1676 (1970), by increasing the federal government’s role in the battle against air pollution. Train v. Natural Resources Defense Council, Inc., 421 U.S. 60, 64, 95 S.Ct. 1470, 43 L.Ed.2d 731 (1975). The amendments require the states to develop pollution control programs (State Implementation Plans or SIPs) that will keep the levels of given pollutants in the atmosphere below the National Ambient Air Quality Standards (NAAQSs) set by EPA. Clean Air Act §§ 109, 110, 42 U.S.C. §§ 1857C-4, 1857c-5 (1970 & Supp V *3221975), as amended, Pub.L.No. 95-95 § 106-108, 91 Stat. 691-697.5

In addition, the 1970 amendments added Section 111, which is the focus of this litigation. This section directs EPA to set specific and rigorous limits on the amounts of pollutants that may be emitted from any “new source” of air pollution. The New Source Performance Standards (NSPSs) established under Section 111 are designed to force new sources to employ the best demonstrated systems of emission reduction.6 Since the NSPSs are likely to be stricter than emission standards under State Implementation Plans, plant operators have an incentive to avoid application of the NSPSs.

The basic controversy in the cases before us concerns the determination of the units to which the NSPSs apply. Under the Act the NSPSs apply to “new sources.” A “new source” is defined as “any stationary source, the construction or modification of which” begins after the NSPS covering that type of source is published. Section 111(a)(2), 42 U.S.C. § 1857c-6(a)(2) (1970) (emphasis added). Further statutory definitions explain the terms used in this one. A “ ‘stationary source’ means any building, structure, facility, or installation which emits or may emit any air pollutant.” Section 111(a)(3), 42 U.S.C. § 1857e-6(a)(3) (1970). A “ ‘modification’ means any physical change in, or change in the method of operation of, a stationary source which increases the amount of any air pollutant emitted by such source or which results in the emission of any air pollutant not previously emitted.” Section 111(a)(4), 42 U.S.C. § 1857c-6(a)(4) (1970). The statute thus directs that the NSPSs are to apply to any building, structure, facility, or installation which emits or may emit any air pollutant and which is either (1) newly constructed or (2) physically or operationally changed in such a way that its emission of any air pollutant increases.7

The “bubble concept” is based on defining a stationary source as a combination of facilities, such as an entire plant, and applying the NSPSs only when a new plant is constructed or when an existing plant is physically or operationally changed in such a way that net emissions of any pollutant from the entire plant increase. If applied consistently, the bubble concept would allow the operator of an existing plant to avoid application of the strict NSPSs by offsetting any increase in pollution caused by a change in the plant (e.g., modification or replacement of an existing facility, or even addition of a new facility) against a decrease in pollution from other units within the plant as a whole.

B. History of the EPA Regulations

EPA’s original regulations interpreting Section 111, promulgated in 1971, repeated *323the statutory definitions of “stationary source” and “modification” almost word for word8 and did not contain any version of the “bubble concept.” See 36 Fed.Reg. 24877 (1971), codified at 40 C.F.R. §§ 60.2(d) & (h) (1975). EPA acknowledges in its brief that it originally “defined the term new source as ‘an affected facility’ which in turn was defined as any apparatus to which a standard of performance is specifically applicable.” Brief for EPA at 10 (emphasis added). Affected facilities, and thus new sources, were clearly not synonymous with entire plants. For example, the regulations setting the NSPSs for sources in Portland cement plants identify the following “affected facilities” in such plants: “kiln, clinker cooler, raw mill system, finish mill system, raw mill dryer, raw material storage, clinker storage, finished product storage, conveyor transfer points, bagging and bulk loading and unloading systems.” 40 C.F.R. § 60.60 (1975).9

The “bubble concept” appeared first in proposals from the nonferrous smelting industry and the Department of Commerce (DOC) beginning in December 1972.10 The industry and DOC urged that a stationary source be defined as an entire plant so that no “modifications” of the source would occur unless the total emissions of some pollutant from the plant increased.

In response to industry proposals and demands from DOC, EPA prepared draft regulations in May and July 1974 “clarifying” the previous regulations.11 EPA made some concessions to the industry in these drafts,12 but did not accept the industry’s position that an entire plant should be defined as a single source.13 EPA contin*324ued to resist the bubble concept in meetings with industry representatives, DOC, and the Office of Management and Budget through August 1974. See JA 8-9, 30-34.

Then, in September 1974, the agency again revised its position, making further concessions and proposing new regulations incorporating a limited version of the bubble concept.14 After an additional concession further extending the bubble concept in response to a submission by DOC,15 the proposed regulations were adopted by EPA.16

The new regulations would classify an entire plant as a single stationary source by embellishing the statutory definition of a stationary source as follows:

“Stationary source” means any building, structure, facility, or installation which emits or may emit any air pollutant and which contains any one or combination of the following:

(1) Affected facilities.

(2) Existing facilities.

(3) Facilities of the type for which no standards have been promulgated in this part.

40 C.F.R. § 60.2(d) (1976) (emphasis added). The italicized language is not included in the statutory definition of “stationary source” (“any building, structure, facility, or installation which emits or may emit any air pollutant”), nor was it included in the prior regulations. See 40 C.F.R. § 60.2(d) (1975). Thus the present regulations, instead of limiting the definition of “stationary source” to one “facility” as the statute does, make it cover “any one or combination of” facilities.17 The preamble to the new regulations makes it clear that the purpose of this change is to define a statutory source as an entire plant.18

*325Relying on this new definition of a statutory source, EPA applies the bubble concept to allow a plant operator who alters an existing facility in a way that increases its emissions to avoid application of the NSPSs by decreasing emissions from other facilities within the plant. The regulations provide that “[a] modification shall not be deemed to occur” unless the change in an existing facility results in a net increase in the emission of a pollutant from the whole “source.”19

In spite of strong comments from the nonferrous smelting industry, represented here by ASARCO,20 the new regulations do not consistently apply the bubble concept to treat an entire plant as a single stationary source. EPA continues to apply NSPSs to all newly constructed facilities, even when the emission increases from the new facilities are offset by emission decreases from other facilities in the same plant. Newly constructed facilities are thus treated as independent stationary sources. In order to draw a line defining when the bubble concept will be applied, the regulations classify any changes in existing facilities that cost more than a fixed percentage of the value of the changed facility as “reconstruction.” 40 C.F.R. § 60.15 (1976). “Reconstructed” facilities, like new facilities, are subject to NSPSs regardless of whether emissions from the plant of which they are a part increase.

In its petition for review ASARCO argues that the bubble concept must be applied to allow emission increases from reconstruction and new construction to be offset. Sierra argues that the Act defines a “source” as an individual facility, as distinguished from a combination of facilities such as a plant, and that the bubble concept must therefore be rejected in toto. For the reasons stated below we agree with Sierra and remand to EPA for further proceedings consistent with this opinion.

II

A. Scope of Review

The proper scope for judicial review of EPA’s regulations interpreting the Clean Air Act is defined in Section 10(e) of the Administrative Procedure Act, which requires a reviewing court to “decide all relevant questions of law, interpret * * statutory provisions, and * * * hold unlawful and set aside agency action, findings, and conclusions found to be * * * in excess of statutory jurisdiction, authority, or limitations, or short of statutory right * * 5 U.S.C. § 706(2)(C) (1970). See Ethyl Corp. v. EPA, 176 U.S.App.D.C. 373, 541 F.2d 1, 33-34 & n.71 (en banc), cert. denied, 426 U.S. 941, 96 S.Ct. 2663, 49 L.Ed.2d 394 (1976). To be sure, as EPA correctly points out, the Supreme Court and this court have both stated that EPA’s interpretation of the Clean Air Act is to be given considerable deference.21 Nevertheless, it is clear that a reviewing court has the responsibility to examine carefully the words of the statute, the legislative history, and the reasons advanced by the agency to justify its interpretation in order to deter*326mine whether the agency’s interpretation is “sufficiently reasonable that it should [be] accepted by the reviewing courts.” Train v. Natural Resources Defense Council, Inc., supra, 421 U.S. at 75, 95 S.Ct. at 1480.22 Indeed, a panel of this court has recently rejected EPA’s interpretation of another provision of the Clean Air Act. Lubrizol Corp. v. EPA, 183 U.S.App.D.C. 288, 562 F.2d 807 (1977) (finding that even the agency’s persuasive policy arguments were inadequate to extend the commonly accepted meaning of the word “fuel” to include “motor oil” in the absence of any significant support for such an extension in the Act’s legislative history). See also Ass’n of American Railroads v. Costle, 183 U.S.App. D.C. 362, 562 F.2d 1310 (1977) (rejecting the EPA Administrator’s contention that the Noise Control Act of 1972 allowed him discretion not to issue federal noise standards for certain “equipment and facilities” on railroads).

Following the approach defined by these precedents, we first consider the challenge to the regulations raised by the Sierra Club, and then dispose of the challenge raised by ASARCO.

B. Sierra’s Challenge to the Regulations

The Sierra Club’s basic contention is that the new regulations are inconsistent with the plain language of Section 111. The statute defines a stationary source as “any building, structure, facility, or installation which emits or may emit any air pollutant.” Section 111(a)(3), 42 U.S.C. § 1857c-6(a)(3) (1970) (emphasis added). In contrast, the new regulations define stationary source to include “any * * * combination of * * facilities * * 40 C.F.R. § 60.2(d) (1976) (emphasis added).

This change in the definition of a stationary source is essential to EPA’s adoption of ' the bubble concept. By treating a combination of facilities as a single source, the regulations allow a facility whose emissions are increased by alterations to avoid complying with the applicable NSPS as long as emission decreases from other facilities within the same “source” cancel out the increase from the altered facility.23 Sierra argues forcefully that this result is incompatible with the statute’s mandate that NSPSs should be applied to “any structure, building, facility, or installation” that undergoes “any physical change * * * or * * * change in the method of operation * * * which increases the amount of any air pollutant emitted by such [structure, building, facility, or installation].” 42 U.S.C. §§ 1857c-6(a)(3), 1857c-6(a)(4) (1970) (emphasis added). See brief for petitioner Sierra Club at 25-33.

EPA responds that the “broad” statutory definition of stationary source gives it “discretion” to define a stationary source as either a single facility or a combination of facilities. Brief for .EPA at 13-16. We find this response unpersuasive.24 The reg*327ulations plainly indicate that EPA has attempted to change the basic unit to which the NSPSs apply from a single building, structure, facility, or installation — the unit prescribed in the statute — to a combination of such units. The agency has no authority to rewrite the statute in this fashion. See, e. g., Ass’n of American Railroads v. Costle, supra; Lubrizol Corp. v. EPA, supra.

Our conclusion that the regulations incorporating the bubble concept must be rejected as inconsistent with the language of the Act is reinforced when we consider the purpose of the Clean Air Act and Section 111, the confusion generated by the present regulations, and the weakness of EPA’s arguments in favor of the bubble concept.

“[T]he goal of the Clean Air Act,” as EPA admits in its brief, “is to enhance air quality and not merely to maintain it.” Brief for EPA at 17 (emphasis added). See Clean Air Act § 101(b)(1), 42 U.S.C. § 1857(b)(1) (1970). Section Ill’s provisions mandating New Source Performance Standards were passed because Congress feared that the system of state plans designed to keep air pollution below nationally determined levels was insufficient by itself to achieve the goal of protecting and improving air quality.25 The New Source Performance Standards are designed to enhance air quality by forcing all newly constructed or modified buildings, structures, facilities, or installations to employ pollution control systems that will limit emissions to the level “achievable through application of the best technological system of continuous emission reduction which * * the Administrator determines has been adequately demonstrated.” 42 U.S.C. § 1857c-6(a)(1) (1970),. as amended, Pub.L.No. 95-95 § 109(c)(1), 91 Stat. 699-700.26 The bubble *328concept in the challenged regulations would undercut Section 111 by allowing operators to avoid installing the best pollution control technology on an altered facility as long as the emissions from the entire plant do not increase. For example, under the bubble concept an operator who alters one of its facilities so that its emission of some pollutant increases might avoid application of the NSPS by simultaneously equipping other plant facilities, with additional, but inferior, pollution control technology or merely reducing their production.27 Applying the bubble concept thus postpones the time when the best technology must be employed and at best maintains the present level of emissions.

Moreover, the challenged regulations are internally inconsistent and create confusion by defining a stationary source one way (as an entire plant) when determining whether a “source” has been “modified,” and another way (as an individual facility) when determining whether a “source” has been newly constructed or “reconstructed.”28 This inconsistency is apparently the result of a “compromise” between EPA’s original regulations, which followed the Act in treating a single facility as a source, and the industry position presented by ASARCO that a source must be defined as an entire plant.29 We are unable to understand why EPA should find it necessary to compromise by adopting a position that it admits is contrary to both the language and the basic purpose of the Act.30

EPA’s main argument in support of its regulations is that its version of the bubble concept is necessary to provide flexibility in applying the NSPSs to modified facilities because the cost of bringing existing facilities into compliance with NSPSs is allegedly much greater than the cost of bringing new facilities into compliance.31 This argument does not survive analysis. The record does not show that any version of the bubble concept is needed to provide flexibility to the operators of existing facilities. Under provisions of the regulations that are not challenged in this litigation, the operator of an existing facility can make any alterations he wishes in the facility without becoming subject to the NSPS as long as the level of emissions from the altered facility does not increase32 Thus the level of emissions before alterations take place, rather than the strict NSPS, effectively defines the standard that an altered facility *329must meet. The record does not indicate why more flexibility than this is necessary or even appropriate. Even if flexibility were a problem, the statute on its face allows for cost considerations to be taken into account in setting NSPSs, rather than in determining whether the standards will apply to whole plants or to individual facilities within those plants.33

Finally, the record indicates that the bubble concept has been supported by examples drawn from circumstances peculiar to the nonferrous smelting industry.34 As EPA itself recognizes in its brief,35 the proper place to consider the problems a particular industry will have in meeting the NSPSs is in proceedings dealing with the standards for that particular industry, not in regulations setting standards for all industries.36

We therefore agree with the Sierra Club that EPA’s regulations incorporating the bubble concept are inconsistent with the language and purpose of the statute and cannot be justified by any alleged need for flexibility.

C. ASARCO's Challenge to the Regulations

The dispute between ASARCO and EPA centers on how far the bubble concept should extend. ASARCO asserts that a stationary source must be defined as an entire plant for all purposes and that the NSPSs should therefore never apply to an existing plant, even if new facilities are built or old ones are “reconstructed,” unless the net emissions of some pollutant from the entire plant increase.37 ASARCO argues that the inconsistency in EPA’s present definition of a stationary source38 — an inconsistency created by EPA’s partial concessions to ASARCO’s position — should be resolved39 by doing away with all limitations on the bubble concept. Brief for petitioner ASARCO at 18-23.

Since we find that any version of the bubble concept is incompatible with the language of the Act and contrary to its purpose, ASARCO’s position is clearly untenable.40 Our holding therefore resolves the *330inconsistency in the challenged regulations by eliminating its source: the bubble concept.

Accordingly, we remand to the EPA for further proceedings not inconsistent with this opinion.

So ordered.

LEVENTHAL, Circuit Judge,

concurring:

I concur in Judge Wright’s opinion for the court.

The flexibility of a concurring opinion1 permits some observations on the flexibility that may be available to the agency.

Conditions in the nonferrous smelting industry apparently prompted the approach reflected in the challenged regulations. The cost of controlling pollution emitted by existing smelters is great. The industry has argued that more efficient pollution control can be achieved by reducing the pollutants emitted by other facilities in the smelting process (a less expensive undertaking) to compensate for any increase in pollution attributable to modified smelters. The agency’s response has been to permit this tactic in the case of all modifications of existing facilities, but not in the case of newly constructed facilities (where the cost of building in appropriate pollution controls may be more modest).

As the majority opinion points out, the manner in which the statute defines the terms “new source” and “stationary source” does not provide leeway for the agency to assign different meanings to “stationary source” depending on whether construction or modification is in question.

However, the statute does not totally prohibit the agency from making appropriate distinctions based on the realistic differences between new construction and modification. While costs may not be considered in determining whether a facility will be' subject to an NSPS, as the bubble concept would have done, they may be considered in determining the level at which a standard should be set, and how it should be formulated.

The statute permits the Administrator to “distinguish among classes, types, and sizes within categories of new sources for the purpose of establishing [NSPS’s].”2 I believe the terms “classes” and “types” are broad enough to permit the Administrator to set more liberal standards for modified facilities than for newly constructed facilities that perform the same function, if he determines that such a distinction is appropriate in a given case. This approach permits custom tailoring, and thus stands in contrast to the challenged regulations.

The challenged regulations potentially immunize all modified facilities — and, as Judge Wright points out, would contravene the policy that contemplated that modification would bring about improvements — and would provide this escape without regard to the costliness of conforming to performance standards for a particular category of modified facilities. The record is devoid of support for the proposition that pollution control for modified facilities will always involve costs disproportionate to the resultant benefits. However, the flexibility to distinguish between classes of new sources may serve to authorize a differential in the stan*331dards applicable to new and modified equipment in those cases where warranted by cost differences and cost-benefit analysis. This approach would not permit the Administrator to immunize a modified facility (one type of new source) from regulation under a performance standard,3 but would permit an alternative course that promotes the underlying statutory concept of progressively bringing all pollution sources within the constraint of performance standards.

MacKINNON, Circuit Judge,

concurring in part and dissenting in part:

I dissent from the judgment of the court for two general reasons. Primarily, I feel that the Administrator is vested with authority — under the terms of 42 U.S.C. § 1857c-6(b)(l)(B) directing him to “establish Federal standards of performance for new sources” and 42 U.S.C. § 1857c-6(b)(2) authorizing him to distinguish among classes, types, and sizes within categories of new sources for the purpose of establishing such standards” — to promulgate regulations effectively exempting certain “modified” stationary sources from the requirement of compliance with New Source Performance Standards (NSPSs). The court’s opinion does not adequately consider the possibility, suggested by Judge Leventhal’s concurrence but not taken to its logical conclusion there, that in stating that certain modifications will not be subject to NSPSs, the Administrator is in fact promulgating an NSPS — eoncededly a somewhat unusual one but nevertheless one fully within the Administrator’s discretion to establish- — for these sources. No reason has been advanced why the “standards” that the EPA is directed to promulgated cannot in certain cases be standards of nonregulation or simply of prior emission levels, and this is all that the Agency has in effect attempted to do in this case.

I

I agree with Judge Leventhal that the Administrator could promulgate different NSPSs for modified as opposed to reconstructed or new stationary sources, but I cannot see why Judge Leventhal — -while conceding that the Administrator has the authority to set such different standards— refuses to accept that the Agency can, in “distinguishing among classes, types and sizes, within categories of new sources,” establish as NSPSs- that certain sources need not conform to any new standards. If the Administrator could set the NSPS so high for modified, as opposed to reconstructed or new, sources that the former would plausibly never violate such standards (or merely establish standards sufficiently different for modified as opposed to new or reconstructed sources that the former would be left significantly less regulated), why should the Agency be disabled from effectively exempting them from NSPSs altogether? Merely because the Administrator has phrased his different treatment for modified as opposed to new or reconstructed sources in terms of an exemption from a standard rather than in terms of a lesser standard we should not determine that he has exceeded his authority. To do so seems overly formalistic in the context of a statute in which there is abundant language confirming that Congress, recognizing the difficulty of balancing economic and ecological goals, had meant to extend generous authority for discretionary decisions to the EPA.

The court argues that — regardless of what other authority Congress meant to grant the Agency to implement the Clean Air Act — as the regulations embracing the bubble concept defined “facility” and “modification” in terms facially conflicting with those used in the statute itself, the Administrator exceeded his authority in promulgating such rules. I agree that the semantic route chosen by the EPA to effect a partial bubble concept was somewhat dubious, but I feel that the court has inadequately appreciated the background of broad administrative discretion against which the Act is set.

*332One need only peruse section 111 of the Act in order to become convinced of the broadness of the discretion with which Congress meant to clothe the Administrator in implementing that section. For example, in promulgating standards of performance for new stationary sources he has discretion (1) to “publish . . . (and ... revise) a list of categories of stationary sources”; (2) to determine if a category of stationary sources “contributes significantly” to air pollution which causes or contributes to the endangerment of public health or welfare; (3) to “promulgate . . . such standards with such modifications as he deems appropriate . . . [and] from time to time, revise such standards . . .” (Emphasis added). These grants of discretionary authority, when read in conjunction with the authority, referred to above, to distinguish between categories of new sources, point so clearly to a congressional intent to allow the Administrator a wide-ranging license to accommodate the environmental mandate of the Act with the exigencies of technology and economics that I am unwilling to accept the court’s decision that the bubble concept lies outside the reach of his discretion.

The misguided literalism of the court’s approach is perhaps better exposed by examining its effect on another regulation, more obviously legitimate than the bubble concept, promulgated by the Administrator. In 40 C.F.R. § 60.14(e)(1) “maintenance, repair, and replacement which the Administrator determines to be routine for a source category” are held not to be “modifications.” Applying the same rigorous semantic technique that the court invokes in proscribing the bubble concept, however, one would be compelled to insist that 42 U.S.C. § 1857c-6(a)(4) defines modification to be “any physical change . . . which increases the amount of air pollutant emitted.” (Emphasis added.) It is hard to see how the court could consistently reject the EPA’s regulations embracing the bubble concept and not also reject its exemption of routine maintenance from the category of “modification” when' the latter obviously involves a “physical change” and may increase pollution. It seems to me, however, incredible that Congress would have meant to dissuade manufacturers from repairing their equipment by the potentially harsh threat of being forced to comply with an NSPSs, or that this court would reject the EPA’s recognized expertise in administering the Clean Air Act and invalidate 40 C.F.R. § 60.14(e)(1). If this court would do so, as its reasoning in this case suggests it would, I fear that it will go far toward preempting the discretion Congress plainly intended to allow the EPA in accommodating the various conflicting interests affected by the Clean Air Act.

In addition to the general grants of discretionary power to the Administrator under the Act, there is a persuasive concrete indication that Congress meant to invest the Administrator with discretion to promulgate the bubble concept, or something very much like it, in the contrast between the prohibitions on modifications under section 111 of the Act (42 U.S.C. § 1857c-6) — with which we are here concerned — and under section 112 (42 U.S.C. § 1857c-7) dealing with hazardous pollutants. Under section 111, it is only unlawful for “an owner or operator of any new source to operate such source in violation of any standard of performance applicable to such source,’’ 42 U.S.C. § 1857c-6(e) (emphasis added), i.e., in violation of the regulation promulgated by the Administrator— which regulation might embrace the bubble concept. In sharp contrast to this relatively mild prohibition is the language of section 112(c) which specifically provides that “no person may . . . modify any existing source which, in the Administrator’s judgment, will emit [a hazardous] air pollutant . in violation of [the national emission] standard [for that pollutant].” I agree that were we concerned here with hazardous air pollutants, the Administrator would have exceeded his authority had he promulgated a bubble concept. However, that Congress clearly knew how to proscribe this concept in section 112 and did not do so in section 111 seems to me to militate against the holding of the court. *333In my view of the statute Congress plainly intended, as one would expect they would, to treat modifications involving hazardous air pollutants more strictly than those involving non-hazardous ones, and yet the court seems willing to ignore the self-evident difference between section 111 and 112 because of a devotion to literalist statutory construction and the truism that the Act was meant not only to maintain, but to enhance air quality. Congress was manifestly aware of how to proscribe the bubble concept when it desired to do so. It did not do so in section 111, and neither should this court.

II

My disagreement with the majority’s proscription of the bubble concept is aggravated by a conviction that in many instances the Administrator has applied too narrow a meaning to the word “facilities” in promulgating his regulations governing emission standards. I doubt, although I need not reach the issue, that the Administrator could be said to have abused his expansive discretion in promulgating these regulations, but in my opinion the narrowness with which he particularized the concept of “facility” in the context of rules for various industries not only distorted legislative purpose but also accentuated the need for something like the bubble concept. The EPA’s attempt to embrace the bubble concept was a laudable attempt to adjust its regulations to be more consistent with the true intent of Congress. It is this corrective effort that the court invalidates.

Congress defined a “stationary source” to mean “any building, structure, facility, or installation which emits or may emit any air pollutant.” 42 U.S.C. § 1857c-6(a)(3) (emphasis added). It did not define “facility,” and specifically did not indicate that it intended a facility to mean anything that may emit any air pollutant. The EPA regulations do not define “facility” either. However, an “affected facility” (emphasis added) is defined in the regulations to mean “with reference to a stationary source, any apparatus to which a standard is applicable.” 40 C.F.R. § 60.2(e). 40 C.F.R. § 60.-2(aa) defining existing facility and the EPA’s application of these terms in various concrete contexts seems similarly restrictive. The regulations demonstrate a distinct reluctance to refer to entire plants as single facilities. This was not an attitude shared by Congress, and certainly not one in accord with common usage. (Of course, if an entire plant is deemed a “facility,” the bubble concept has in many cases been effectively embraced, as then, by the very statutory language on which the court’s opinion relies, any of the plant’s constituent parts could be modified although their particular emission level was increased so long as that of the plant in general remained constant).

In ascertaining the statutory meaning of the term “facility,” it is important to note that Congress used the word in conjunction with “building, structure . . . [and] installation^].” These terms, with which “facility” is grouped, all refer to conglomerate entities capable of housing a number of machines, pieces of equipment, etc. Even if the ordinary usage of the term did not persuade that Congress intended it to comprehend entire plants, or at least such large and independent devices as, for example, blast furnaces, the context in which “facility” is used in the statute would go far towards convincing us of this interpretation, and accordingly of the legislative sanction for the bubble concept.

Moreover, the common meaning of “facility” is such that Congress would have to have shown some affirmative desire to restrict this meaning to preclude “facility” encompassing, in some instances, whole plants. “Facility” does not have any precise meaning, nor is it a legal term of art; its meaning must be determined from common usage, see e.g., People ex rel. Schlaeger v. Bunge Bros. Coal Co., 392 Ill. 153, 64 N.E.2d 365 (1946); State ex rel. Knight v. Cave, 20 Mont. 468, 52 P. 200 (1898), and here in part from its statutory context. The common usage of the word is set forth in Webster’s Third International Dictionary 812-813 (1961) which defines “facility” as *334“something (as a hospital, machinery, plumbing) that is built, constructed, installed or established to perform some particular function or to serve or facilitate some particular end.” (Emphasis added.) A hospital seems an appropriate equivalent of a “plant,” and “facility” is often used to refer to the latter, e.g., Falkner v. Northern States Power Co., 75 Wisc.2d 116, 248 N.W.2d 885, 901 (1977) (nuclear power station is a “facility”); Lincoln Bank & Trust Co. v. Exchange National Bank & Trust Co., Ardmore, Okla., 383 F.2d 694, 697 (10th Cir. 1967) (bank is “facility”).

“Facility” does not refer to a small machine or piece of equipment, but only to those that are of sufficient magnitude and functional independence to perform a separable, integrated operation. For example, a taconite plant consists of a large grinding machine (one machine), the extracting operation (many machines), and the pelletizing operation (two coordinated machines). Collectively, the combination of these three operations constitute a facility, and if the three operations were housed separately each might be termed a facility, but it would not be in keeping with the meaning of the word as used in the statute to refer to each of the many machines in the pelle-tizing operation, as a “facility.” In combination they are, separately they are not.

There is no indication that in drafting the Clean Air Act Congress intended to use “facility” as a term of art, more restricted in scope than it is in ordinary parlance; indeed, as mentioned above, the terms with which the Act groups “facilities” indicates quite the contrary. It should be noted, moreover, that the statute nowhere distinguishes facility and plant. All we can conclude from the phrasing of the Act is that Congress considered “building,” “structure” and “installation” to have somewhat similar connotations but some differences from a “facility,” and this alone is not at all suggestive that the legislature meant to treat individual machines as facilities or that it meant to distinguish facilities and plants.

The regulations, however, while they apply “facility” in its customary — and statutory — sense in certain places, e.g., Iron and Steel Plants (oxygen process furnace) § 60.-140; Coal Preparation Plant, § 60.215(a); Asphalt Concrete Plant, § 60.90; Incinerators of stated consumption, § 60.50, also contain numerous instances where the “apparatus” the regulations sometimes refer to as a “facility” do not rise to the dimensions of a “facility” as that word is used in the statute. Under my interpretation of the regulations, there are numerous instances where the Administrator has referred to individual units of machines, equipment and apparatus as “facilities” where such equipment is not of sufficient size, quantity or capacity to constitute a “facility” as the statute uses that term.

For example, in portland cement factories, the following are deemed “facilities”: kiln, clinker cooler, raw mill system, finish mill system, raw mill dryer, raw material storage, clinker storage, finished product storage, conveyor transfer points, bagging and bulk loading and unloading systems, 40 C.F.R. § 60.60 (1976); and in primary lead smelters, the following are deemed “facilities”: sintering machine, sintering machine discharge end, blast furnace, dross rever-beratory furnace, electric smelting furnace, and converter. 40 C.F.R. § 60.180 (1976). While some of these “facilities” — e.g., the blast furnace in the primary lead smelters — may indeed correspond to the statutory meaning of the term, the tendency to splinter functioning wholes into smaller “apparatuses” seems clear.

Because the Administrator often defines “facility" too narrowly for purposes of the regulations, the bubble concept serves a valuable corrective function. This concept places the restriction that no modifications increasing air pollution be made not on what the Administrator has established in the first instance to be a facility, but on a larger unit. Where it is this larger unit that more nearly approaches the statutory meaning of facility, the bubble concept has effectively served to bring the regulations more in line with the legislative intent. As I feel that Congress intended to deal, in most cases, with entire plants as facilities *335under section 111, the bubble concept — although a more circuitous approach than simply revising the regulations establishing which apparatuses are “facilities” — is not only within the Administrator’s discretion, but also an accomplishment of congressional purpose.

The majority opinion, in sharp contrast to the analysis set forth above, appears to dismiss out of hand any suggestion that “facility” could refer to an entire plant:

Thus, the present regulations, instead of limiting the definition of “stationary source” to one “facility” as the statute does, make it cover “any one or combination of” facilities. The preamble to the new regulations makes it clear that the purpose of this change is to define a statutory source as an entire plant.

188 U.S.App.D.C. at pp.---, 578 F.2d at p. 324 (footnotes omitted). This overly narrow application of the term “facility” leads to an unwarranted antipathy to the bubble concept. Quite apart from the issue of the range or discretion granted the Administrator, because the majority opinion apparently sees facilities in terms of individual machines rather than integrated combinations thereof, it is forced to perceive the bubble concept as a perversion rather than a fulfillment of the legislative objectives. And because the majority opinion sees the concept as a distortion of the statute, it cannot be receptive to the argument that although it might have been more direct for the Administrator simply to re-promulgate his regulations applying the term “facilities” in various specific contexts, the Agency’s adoption of the bubble concept was a realization, not a distortion, of the dictates of the Clean Air Act.

To summarize, in those cases where the bubble concept was only serving to offset the existing regulations’ restrictive interpretation of “facility,” promulgating the regulations embodying this concept was not only a matter within the Administrator’s discretion, it was fully consonant with the terms of the statute. In those cases where the regulations’ application of “facility” to particular industries is in accord with the usual and statutory meaning of the term, the bubble concept — although it may in certain instances be somewhat at variance with the statutory language — is within the broad discretion vested in the Administrator. In the latter cases the majority opinion significantly reduces the flexibility with which the Act was intended to be implemented; in the former cases, the opinion misapprehends the words of the statute.

Ill

The court’s failure to perceive the over-narrowness of its, and in some instances the Administrator’s, interpretation of “facility”; and its apparent refusal to appreciate the extremely broad discretion with which the EPA was vested in order to reconcile the various vital and contradictory interests profoundly affected by the implementation of the Act in my opinion is due, at least in part, to the somewhat abstract terms in which the instant dispute was presented. Constitutional considerations aside, when courts attempt to adjudicate under such circumstances as we have here they operate, lacking the concreteness of a particularized dispute, in an informational vacuum. As a result, the tribunal is liable to err due to the necessity of evaluating the issues before it in the abstract rather than through the perspective gained from examining an actual instance in which the issues arise where the regulations are actually being applied.

While I do not feel that this case should be dismissed either for prematurity or lack of standing, Abbott Laboratories v. Gardner, 387 U.S. 136, 87 S.Ct. 1507, 18 L.Ed.2d 681 (1967), I do feel that it has many of the earmarks of cases that are disposed of on such grounds. The law of standing is proverbially confused, see, e.g., K. Davis, Administrative Law Text § 22.01 (3d ed. 1972); Scott, Standing in the Supreme Court — A Functional Analysis, 86 Harv.L.Rev. 645, 645 n.1 (1973); Lewis, Constitutional Rights and the Misuse of “Standing,” 14 Stan.L. Rev. 433, 434; Note, Standing to Assert Constitutional Jus Tertii, 88 Harv.L.Rev. 423, 423 n.1 (1974), and recent Supreme *336Court cases seem to adopt virtually contradictory approaches towards it, compare Planned Parenthood of Central Missouri v. Danforth, 428 U.S. 52, 96 S.Ct. 2831, 49 L.Ed.2d 788 (1976) and Linda B. S. v. Richard D., 410 U.S. 614, 93 S.Ct. 1146, 35 L.Ed.2d 536 (1973). At present, the law appears to be retreating from the generous standing doctrine of several years past, compare United States v. SCRAP, 412 U.S. 669, 93 S.Ct. 2405, 37 L.Ed.2d 254 (1973) and Warth v. Seldin, 422 U.S. 490, 95 S.Ct. 2197, 45 L.Ed.2d 343 (1975). A “trifling or speculative” interest is no longer sufficient to admit the plaintiff to the courtroom, see Warth v. Seldin, supra; compare United States v. SCRAP, supra; Linda R. S. v. Richard D., supra; Trafficante v. Metropolitan Life Insurance Co., 409 U.S. 205, 211-212, 93 S.Ct. 364, 34 L.Ed.2d 415 (1972). Current doctrine indicates that one who is regulated by a statute or regulation does not ipso facto have standing to challenge it, but rather “the law is that one who is regulated lacks standing unless he can show an interest that is adversely affected” (emphasis added), K. Davis, Administrative Law in the Seventies, Appendix 177 (1977). For example, in Warth v. Seldin, supra —the latest major Supreme Court ruling on standing — builders governed by an ordinance were denied standing even though they alleged that the statute prevented them from building the type of housing they desired, as were other low income plaintiffs who alleged that they desired to obtain such housing. In that case, the Court stated:

Petitioners must allege and show that they personally have been injured, not that injury has been suffered by other, unidentified members of the class to which they belong and which they purport to represent. . . We hold only that a plaintiff who seeks to challenge exclusionary zoning practices must allege specific, concrete facts demonstrating that the challenged practices harm him, and that he personally would benefit in a tangible way from the court’s intervention. Absent the necessary allegations of demonstrable, particularized injury, there can be no confidence of “a real need to exercise the power of judicial review” or that relief can be framed “no [broader] than required by the precise facts to which the court’s ruling would be applied.”

422 U.S. at 502, 508, 95 S.Ct. at 2210.

The Sierra Club’s interest in this case is very speculative. Allegation of injury to one’s environment is sufficient to have standing, but there has as yet been no tangible injury to the environment. Cf. Sierra Club v. Morton, 405 U.S. 727, 92 S.Ct. 1361, 31 L.Ed.2d 636 (1972), and the possible deleterious effect of the EPA’s accommodation of the industry’s desire for the “bubble concept” are far from clear. Perhaps the flexibility which this concept would allow manufacturers in developing and installing new machinery now could in the future, when new means of pollution control are found, actually allow for quicker enhancement of air quality than would have been possible had the bubble concept been totally repudiated. If and when the Sierra Club can demonstrate a concrete injury, they may, of course, gain access to the courts, see Sierra Club v. Morton, 348 F.Supp. 219 (N.D.Cal. 1972), but at this point there must be considerable doubt as to whether they are legitimate plaintiffs.

The question of standing naturally suggests the issue of ripeness, see K. Davis, Administrative Law Text § 22.01 (1958). It has long been clear that courts will not— whether for constitutional or prudential reasons is a matter of some debate, see e.g., Berger, Standing to Sue in Public Actions: Is It a Constitutional Right, 79 Yale L.J. 816 (1969) — adjudicate a “controversy” which has not become sufficiently concrete, see e.g., Rescue Army v. Municipal Court, 331 U.S. 549, 67 S.Ct. 1409, 91 L.Ed. 1666 (1947); Massachusetts v. Mellon, 262 U.S. 447, 43 S.Ct. 597, 67 L.Ed. 1078 (1923). Absent a particular application of a regulation, this court itself has held that the question was not ripe for adjudication, National Assoc. of Insurance Agents, Inc. v. Board of Governors of the Federal Reserve System, 160 U.S.App.D.C. 144, 489 F.2d 1268, 1271 *337(1974); cf. Alabama Assoc. of Insurance Agents v. Board of Governors of the Federal Reserve System, 533 F.2d 224 (5th Cir. 1976).

To the extent that this case presents the issue of whether agency regulations comply with the statute under which they have been promulgated, it poses a “purely legal” question that is ripe for decision. Abbott Laboratories v. Gardner, supra; Bethlehem Steel v. EPA, 536 F.2d 156 (7th Cir. 1976); National Automatic Laundry & Cleaning Council v. Shultz, 143 U.S.App.D.C. 274, 443 F.2d 689, 695 (1971). This appeal, however, hinges in large part on the proper interpretation of “facility” as used in the statute, and the court in dealing with this issue is to some extent rendering a purely advisory opinion. Had the court been presented with a specific factory (or part thereof), plant, or other constructed unit and been asked to determine whether or not it was a “facility” within the meaning of the Act, it would have been operating within traditional judicial competence, but no such concrete example was presented. Our decision should await such a presentation. The case may not be premature in a formal sense, but the court has indirectly been asked to elucidate the meaning of statutory language and virtually to become a part of the rule-making process, without the benefit of a specific example. I venture to suggest that in this case the court has suffered exactly that “entangling [of] themselves in abstract disagreements over administrative policies” that the ripeness doctrine was meant to avoid, Abbott Laboratories v. Gardner, supra, 387 U.S. at 148, 87 S.Ct. 1507. If a concrete example of a “facility” had been actually involved in this case, the opinion might well have arrived at a different interpretation of the meaning of the term, and one which, in my view, would more closely approximate its meaning in eommonsense and in the statute.

I respectfully dissent to the extent indicated above. As I feel that the Administrator was well within his discretion in applying the bubble concept only to modified and not to new or reconstructed facilities, however, I concur in the result — although not in the rationale — of the court’s dismissal of ASARCO’s argument that the inconsistency of the Administrator’s position required extending the bubble concept to new and reconstructed sources as well as those that had merely been modified.

Justice Stevens

delivered the opinion of the Court.

In the Clean Air Act Amendments of 1977, Pub. L. 95-95, 91 Stat. 685, Congress enacted certain requirements appli*840cable to States that had not achieved the national air quality standards established by the Environmental Protection Agency (EPA) pursuant to earlier legislation. The amended Clean Air Act required these “nonattainment” States to establish a permit program regulating “new or modified major stationary sources” of air pollution. Generally, a permit may not be issued for a new or modified major stationary source unless several stringent conditions are met.1 The EPA regulation promulgated to implement this permit requirement allows a State to adopt a plantwide definition of the term “stationary source.”2 Under this definition, an existing plant that contains several pollution-emitting devices may install or modify one piece of equipment without meeting the permit conditions if the alteration will not increase the total emissions from the plant. The question presented by these cases is whether EPA’s decision to allow States to treat all of the pollution-emitting devices within the same industrial grouping as though they were encased within a single “bubble” is based on a reasonable construction of the statutory term “stationary source.”

I

The EPA regulations containing the plantwide definition of the term stationary source were promulgated on October *84114, 1981. 46 Fed. Reg. 50766. Respondents3 filed a timely petition for review in the United States Court of Appeals for the District of Columbia Circuit pursuant to 42 U. S. C. 17607(b)(1).4 The Court of Appeals set aside the regulations. National Resources Defense Council, Inc. v. Gorsuch, 222 U. S. App. D. C. 268, 685 F. 2d 718 (1982).

The court observed that the relevant part of the amended Clean Air Act “does not explicitly define what Congress envisioned as a ‘stationary source, to which the permit program . . . should apply,” and further stated that the precise issue was not “squarely addressed in the legislative history.” Id., at 273, 685 F. 2d, at 723. In light of its conclusion that the legislative history bearing on the question was “at best contradictory,” it reasoned that “the purposes of the non-attainment program should guide our decision here.” Id., at 276, n. 39, 685 F. 2d, at 726, n. 39.5 Based on two of its precedents concerning the applicability of the bubble concept to certain Clean Air Act programs,6 the court stated that the bubble concept was “mandatory” in programs designed merely to maintain existing air quality, but held that it was “inappropriate” in programs enacted to improve air quality. Id., at 276, 685 F. 2d, at 726. Since the purpose of the per*842mit program — its “raison d’Ure,” in the court’s view — was to improve air quality, the court held that the bubble concept was inapplicable in these cases under its prior precedents. Ibid. It therefore set aside the regulations embodying the bubble concept as contrary to law. We granted certiorari to review that judgment, 461 U. S. 956 (1983), and we now reverse.

The basic legal error of the Court of Appeals was to adopt a static judicial definition of the term “stationary source” when it had decided that Congress itself had not commanded that definition. Respondents do not defend the legal reasoning of the Court of Appeals.7 Nevertheless, since this Court reviews judgments, not opinions,8 we must determine whether the Court of Appeals’ legal error resulted in an erroneous judgment on the validity of the regulations.

HH H-(

When a court reviews an agency s construction of the statute which it administers, it is confronted with two questions. First, always, is the question whether Congress has directly spoken to the precise question at issue. If the intent of Congress is clear, that is the end of the matter; for the court, *843as well as the agency, must give effect to the unambiguously expressed intent of Congress.9 If, however, the court determines Congress has not directly addressed the precise question at issue, the court does not simply impose its own construction on the statute,10 as would be necessary in the absence of an administrative interpretation. Rather, if the statute is silent or ambiguous with respect to the specific issue, the question for the court is whether the agency’s answer is based on a permissible construction of the statute.11

“The power of an administrative agency to administer a congressionally created . . . program necessarily requires the formulation of policy and the making of rules to fill any gap left, implicitly or explicitly, by Congress.” Morton v. Ruiz, 415 U. S. 199, 231 (1974). If Congress has explicitly left a gap for the agency to fill, there is an express delegation *844of authority to the agency to elucidate a specific provision of the statute by regulation. Such legislative regulations are given controlling weight unless they are arbitrary, capricious, or manifestly contrary to the statute.12 Sometimes the legislative delegation to an agency on a particular question is implicit rather than explicit. In such a case, a court may not substitute its own construction of a statutory provison for a reasonable interpretation made by the administrator of an agency.13

We have long recognized that considerable weight should be accorded to an executive department’s construction of a statutory scheme it is entrusted to administer,14 and the principle of deference to administrative interpretations

“has been consistently followed by this Court whenever decision as to the meaning or reach of a statute has involved reconciling conflicting policies, and a full understanding of the force of the statutory policy in the given situation has depended upon more than ordinary knowledge respecting the matters subjected to agency regulations. See, e. g., National Broadcasting Co. v. United States, 319 U. S. 190; Labor Board v. Hearst Publications, Inc., 322 U. S. 111; Republic Aviation Corp. v. *845Labor Board, 324 U. S. 793; Securities & Exchange Comm’n v. Chenery Corp., 332 U. S. 194; Labor Board v. Seven-Up Bottling Co., 344 U. S. 344.

. . If this choice represents a reasonable accommodation of conflicting policies that were committed to the agency’s care by the statute, we should not disturb it unless it appears from the statute or its legislative history that the accommodation is not one that Congress would have sanctioned.” United States v. Shimer, 367 U. S. 374, 382, 383 (1961).

Accord, Capital Cities Cable, Inc. v. Crisp, ante, at 699-700.

In light of these well-settled principles it is clear that the Court of Appeals misconceived the nature of its role in reviewing the regulations at issue. Once it determined, after its own examination of the legislation, that Congress did not actually have an intent regarding the applicability of the bubble concept to the permit program, the question before it was not whether in its view the concept is “inappropriate” in the general context of a program designed to improve air quality, but whether the Administrator’s view that it is appropriate in the context of this particular program is a reasonable one. Based on the examination of the legislation and its history which follows, we agree with the Court of Appeals that Congress did not have a specific intention on the applicability of the bubble concept in these cases, and conclude that the EPA’s use of that concept here is a reasonable policy choice for the agency to make.

Ill

In the 1950’s and the 1960’s Congress enacted a series of statutes designed to encourage and to assist the States in curtailing air pollution. See generally Train v. Natural Resources Defense Council, Inc., 421 U. S. 60, 63-64 (1975). The Clean Air Amendments of 1970, Pub. L. 91-604, 84 Stat. 1676, “sharply increased federal authority and responsibility *846in the continuing effort to combat air pollution,” 421 U. S., at 64, but continued to assign “primary responsibility for assuring air quality” to the several States, 84 Stat. 1678. Section 109 of the 1970 Amendments directed the EPA to promulgate National Ambient Air Quality Standards (NAAQS’s)15 and § 110 directed the States to develop plans (SIP’s) to implement the standards within specified deadlines. In addition, § 111 provided that major new sources of pollution would be required to conform to technology-based, performance standards; the EPA was directed to publish a list of categories of sources of pollution and to establish new source performance standards (NSPS) for each. Section 111(e) prohibited the operation of any new source in violation of a performance standard.

Section 111(a) defined the terms that are to be used in setting and enforcing standards of performance for new stationary sources. It provided:

“For purposes of this section:

“(3) The term ‘stationary source’ means any building, structure, facility, or installation which emits or may emit any air pollutant.” 84 Stat. 1683.

In the 1970 Amendments that definition was not only applicable to the NSPS program required by § 111, but also was made applicable to a requirement of § 110 that each state implementation plan contain a procedure for reviewing the location of any proposed new source and preventing its construction if it would preclude the attainment or maintenance of national air quality standards.16

In due course, the EPA promulgated NAAQS’s, approved SIP’s, and adopted detailed regulations governing NSPS’s *847for various categories of equipment. In one of its programs, the EPA used a plantwide definition of the term “stationary source.” In 1974, it issued NSPS’s for the nonferrous smelting industry that provided that the standards would not apply to the modification of major smelting units if their increased emissions were offset by reductions in other portions of the same plant.17

Nonattainment

The 1970 legislation provided for the attainment of primary NAAQS’s by 1975. In many areas of the country, particularly the most industrialized States, the statutory goals were not attained.18 In 1976, the 94th Congress was confronted with this fundamental problem, as well as many others respecting pollution control. As always in this area, the legislative struggle was basically between interests seeking strict schemes to reduce pollution rapidly to eliminate its social costs and interests advancing the economic concern that strict schemes would retard industrial development with attendant social costs. The 94th Congress, confronting these competing interests, was unable to agree on what response was in the public interest: legislative proposals to deal with nonattainment failed to command the necessary consensus.19

In light of this situation, the EPA published an Emissions Offset Interpretative Ruling in December 1976, see 41 Fed. Reg. 55524, to “fill the gap,” as respondents put it, until Congress acted. The Ruling stated that it was intended to *848address “the issue of whether and to what extent national air quality standards established under the Clean Air Act may restrict or prohibit growth of major new or expanded stationary air pollution sources.” Id., at 55524-55525. In general, the Ruling provided that “a major new source may locate in an area with air quality worse than a national standard only if stringent conditions can be met.” Id., at 55525. The Ruling gave primary emphasis to the rapid attainment of the statute’s environmental goals.20 Consistent with that emphasis, the construction of every new source in nonattainment areas had to meet the “lowest achievable emission rate” under the current state of the art for that type of facility. See Ibid. The 1976 Ruling did not, however, explicitly adopt or reject the “bubble concept.”21

IV

The Clean Air Act Amendments of 1977 are a lengthy, detailed, technical, complex, and comprehensive response to a major social issue. A small portion of the statute — 91 Stat. *849745-751 (Part D of Title I of the amended Act, 42 U. S. C. §§7501-7508) — expressly deals with nonattainment areas. The focal point of this controversy is one phrase in that portion of the Amendments.22

Basically, the statute required each State in a non-attainment area to prepare and obtain approval of a new SIP by July 1, 1979. In the interim those States were required to comply with the EPA’s interpretative Ruling of December 21, 1976. 91 Stat. 745. The deadline for attainment of the primary NAAQS’s was extended until December 31, 1982, and in some cases until December 31, 1987, but the SIP’s were required to contain a number of provisions designed to achieve the goals as expeditiously as possible.23

*850Most significantly for our purposes, the statute provided that each plan shall

“(6) require permits for the construction and operation of new or modified major stationary sources in accordance with section 173 . . . Id., at 747.

Before issuing a permit, § 173 requires (1) the state agency to determine that there will be sufficient emissions reductions in the region to offset the emissions from the new source and also to allow for reasonable further progress toward attainment, or that the increased emissions will not exceed an allowance for growth established pursuant to § 172(b)(5); (2) the applicant to certify that his other sources in the State are in compliance with the SIP, (3) the agency to determine that the applicable SIP is otherwise being implemented, and (4) the proposed source to comply with the lowest achievable emission rate (LAER).24

*851The 1977 Amendments contain no specific reference to the “bubble concept.” Nor do they contain a specific definition of the term “stationary source,” though they did not disturb the definition of “stationary source” contained in § 111(a)(3), applicable by the terms of the Act to the NSPS program. Section 302(j), however, defines the term “major stationary source” as follows:

“(j) Except as otherwise expressly provided, the terms ‘major stationary source’ and ‘major emitting facility’ mean any stationary facility or source of air pollutants which directly emits, or has the potential to emit, one hundred tons per year or more of any air pollutant (including any major emitting facility or source of fugitive emissions of any such pollutant, as determined by rule by the Administrator).” 91 Stat. 770.

V

The legislative history of the portion of the 1977 Amendments dealing with nonattainment areas does not contain any specific comment on the “bubble concept” or the question whether a plantwide definition of a stationary source is permissible under the permit program. It does, however, plainly disclose that in the permit program Congress sought to accommodate the conflict between the economic interest in permitting capital improvements to continue and the environmental interest in improving air quality. Indeed, the House Committee Report identified the economic interest as one of the “two main purposes” of this section of the bill. It stated:

“Section 117 of the bill, adopted during full committee markup establishes a new section 127 of the Clean Air Act. The section has two main purposes: (1) to allow reasonable economic growth to continue in an area while making reasonable further progress to assure attainment of the standards by a fixed date; and (2) to allow *852States greater flexibility for the former purpose than EPA’s present interpretative regulations afford.

“The new provision allows States with nonattainment areas to pursue one of two options. First, the State may proceed under EPA’s present ‘tradeoff’ or ‘offset’ ruling. The Administrator is authorized, moreover, to modify or amend that ruling in accordance with the intent and purposes of this section.

“The State’s second option would be to revise its implementation plan in accordance with this new provision.” H. R. Rep. No. 95-294, p. 211 (1977).25

The portion of the Senate Committee Report dealing with nonattainment areas states generally that it was intended to “supersede the EPA administrative approach,” and that expansion should be permitted if a State could “demonstrate that these facilities can be accommodated within its overall plan to provide for attainment of air quality standards.” S. Rep. No. 95-127, p. 55 (1977). The Senate Report notes the value of “case-by-case review of each new or modified major source of pollution that seeks to locate in a region exceeding an ambient standard,” explaining that such a review “requires matching reductions from existing sources against *853emissions expected from the new source in order to assure that introduction of the new source will not prevent attainment of the applicable standard by the statutory deadline.” Ibid. This description of a case-by-case approach to plant additions, which emphasizes the net consequences of the construction or modification of a new source, as well as its impact on the overall achievement of the national standards, was not, however, addressed to the precise issue raised by these cases.

Senator Mustíe made the following remarks:

“I should note that the test for determining whether a new or modified source is subject to the EPA interpretative regulation [the Offset Ruling] — and to the permit requirements of the revised implementation plans under the conference bill — is whether the source will emit a pollutant into an area which is exceeding a national ambient air quality standard for that pollutant — or precursor. Thus, a new source is still subject to such requirements as ‘lowest achievable emission rate’ even if it is constructed as a replacement for an older facility resulting in a net reduction from previous emission levels.

“A source — including an existing facility ordered to convert to coal — is subject to all the nonattainment requirements as a modified source if it makes any physical change which increases the amount of any air pollutant for which the standards in the area are exceeded.” 123 Cong. Rec. 26847 (1977).

I — I >

As previously noted, prior to the 1977 Amendments, the EPA had adhered to a plantwide definition of the term “source” under a NSPS program. After adoption of the 1977 Amendments, proposals for a plantwide definition were considered in at least three formal proceedings.

In January 1979, the EPA considered the question whether the same restriction on new construction in nonattainment areas that had been included in its December 1976 Ruling *854should be required in the revised SIP’s that were scheduled to go into effect in July 1979. After noting that the 1976 Ruling was ambiguous on the question “whether a plant with a number of different processes and emission.points would be considered a single source,” 44 Fed. Reg. 3276 (1979), the EPA, in effect, provided a bifurcated answer to that question. In those areas that did not have a revised SIP in effect by July 1979, the EPA rejected the plantwide definition; on the other hand, it expressly concluded that the plantwide approach would be permissible in certain circumstances if authorized by an approved SIP. It stated:

“Where a state implementation plan is revised and implemented to satisfy the requirements of Part D, including the reasonable further progress requirement, the plan requirements for major modifications may exempt modifications of existing facilities that are accompanied by intrasource offsets so that there is no net increase in emissions. The agency endorses such exemptions, which would provide greater flexibility to sources to effectively manage their air emissions at least cost.” Ibid.26

*855In April, and again in September 1979', the EPA published additional comments in which it indicated that revised SIP’s could adopt the plantwide definition of source in non-attainment areas in certain circumstances. See id., at 20372, 20379, 51924, 51951, 51958. On the latter occasion, the EPA made a formal rulemaking proposal that would have permitted the use of the “bubble concept” for new installations within a plant as well as for modifications of existing units. It explained:

“‘Bubble’ Exemption: The use of offsets inside the same source is called the ‘bubble.’ EPA proposes use of the definition of ‘source’ (see above) to limit the use of the bubble under nonattainment requirements in the following respects:

“i. Part D SIPs that include all requirements needed to assure reasonable further progress and attainment by the deadline under section 172 and that are being carried out need not restrict the use of a plantwide bubble, the same as under the PSD proposal.

“ii. Part D SIPs that do not meet the requirements specified must limit use of the bubble by including a definition of ‘installation’ as an identifiable piece of process equipment.”27

*856Significantly, the EPA expressly noted that the word “source” might be given a plantwide definition for some purposes and a narrower definition for other purposes. It wrote:

“Source means any building structure, facility, or installation which emits or may emit any regulated pollutant. ‘Building, structure, facility or installation’ means plant in PSD areas and in nonattainment areas except where the growth prohibitions would apply or where no adequate SIP exists or is being carried out.” Id., at 51925.28

The EPA’s summary of its proposed Ruling discloses a flexible rather than rigid definition of the term “source” to implement various policies and programs:

“In summary, EPA is proposing two different ways to define source for different kinds of NSR programs:

“(1) For PSD and complete Part D SIPs, review would apply only to plants, with an unrestricted plant-wide bubble.

“(2) For the offset ruling, restrictions on construction, and incomplete Part D SIPs, review would apply to both plants and individual pieces of process equipment, causing the plant-wide bubble not to apply for new and modified major pieces of equipment.

“In addition, for the restrictions on construction, EPA is proposing to define ‘major modification’ so as to prohibit the bubble entirely. Finally, an alternative discussed but not favored is to have only pieces of process equipment reviewed, resulting in no plant-wide bubble and allowing minor pieces of equipment to escape NSR *857regardless of whether they are within a major plant.” Id., at 51934.

In August 1980, however, the EPA adopted a regulation that, in essence, applied the basic reasoning of the Court of Appeals in these cases. The EPA took particular note of the two then-recent Court of Appeals decisions, which had created the bright-line rule that the “bubble concept” should be employed in a program designed to maintain air quality but not in one designed to enhance air quality. Relying heavily on those cases,29 EPA adopted a dual definition of “source” for nonattainment areas that required a permit whenever a change in either the entire plant, or one of its components, would result in a significant increase in emissions even if the increase was completely offset by reductions elsewhere in the plant. The EPA expressed the opinion that this interpretation was “more consistent with congressional intent” than the plantwide definition because it “would bring in more sources or modifications for review,” 45 Fed. Reg. 52697 (1980), but its primary legal analysis was predicated on the two Court of Appeals decisions.

In 1981 a new administration took office and initiated a “Government-wide reexamination of regulatory burdens and complexities.” 46 Fed. Reg. 16281. In the context of that *858review, the EPA reevaluated the various arguments that had been advanced in connection with the proper definition of the term “source” and concluded that the term should be given the same definition in both nonattainment areas and PSD areas.

In explaining its conclusion, the EPA first noted that the definitional issue was not squarely addressed in either the statute or its legislative history and therefore that the issue involved an agency “judgment as how to best carry out the Act.” Ibid. It then set forth several reasons for concluding that the plantwide definition was more appropriate. It pointed out that the dual definition “can act as a disincentive to new investment and modernization by discouraging modifications to existing facilities” and “can actually retard progress in air pollution control by discouraging replacement of older, dirtier processes or pieces of equipment with new, cleaner ones.” Ibid. Moreover, the new definition “would simplify EPA’s rules by using the same definition of ‘source’ for PSD, nonattainment new source review and the construction moratorium. This reduces confusion and inconsistency. ” Ibid. Finally, the agency explained that additional requirements that remained in place would accomplish the fundamental purposes of achieving attainment with NAAQS’s as expeditiously as possible.30 These conclusions were ex*859pressed in a proposed rulemaking in August 1981 that was formally promulgated in October. See id., at 50766.

VII

In this Court respondents expressly reject the basic rationale of the Court of Appeals’ decision. That court viewed the statutory definition of the term “source” as sufficiently flexible to cover either a plantwide definition, a narrower definition covering each unit within a plant, or a dual definition that could apply to both the entire “bubble” and its components. It interpreted the policies of the statute, however, to mandate the plantwide definition in programs designed to maintain clean air and to forbid it in programs designed to improve air quality. Respondents place a fundamentally different construction on the statute. They contend that the text of thé Act requires the EPA to use a dual definition — if either a component of a plant, or the plant as a whole, emits over 100 tons of pollutant, it is a major stationary source. They thus contend that the EPA rules adopted in 1980, insofar as they apply to the maintenance of the quality of clean air, as well as the 1981 rules which apply to nonattainment areas, violate the statute.31

Statutory Language

The definition of the term “stationary source” in § 111(a)(3) refers to “any building, structure, facility, or installation” which emits air pollution. See supra, at 846. This definition is applicable only to the NSPS program by the express terms of the statute; the text of the statute does not make this defi*860nition applicable to the permit program. Petitioners therefore maintain that there is no statutory language even relevant to ascertaining the meaning of stationary source in the permit program aside from §302(j), which defines the term “major stationary source.” See supra, at 851. We disagree with petitioners on this point.

The definition in §302(j) tells us what the word “major” means — a source must emit at least 100 tons of pollution to qualify — but it sheds virtually no light on the meaning of the term “stationary source.” It does equate a source with a facility — a “major emitting facility” and a “major stationary source” are synonymous under § 302(j). The ordinary meaning of the term “facility” is some collection of integrated elements which has been designed and constructed to achieve some purpose. Moreover, it is certainly no affront to common English usage to take a reference to a major facility or a major source to connote an entire plant as opposed to its constituent parts. Basically, however, the language of §302(j) simply does not compel any given interpretation of the term “source.”

Respondents recognize that, and hence point to § 111(a)(3). Although the definition in that section is not literally applicable to the permit program, it sheds as much light on the meaning of the word “source” as anything in the statute.32 As respondents point out, use of the words “building, structure, facility, or installation,” as the definition of source, could be read to impose the permit conditions on an individual building that is a part of a plant.33 A “word may have a character of its own not to be submerged by its association.” Russell Motor Car Co. v. United States, 261 U. S. 514, 519 *861(1923). On the other hand, the meaning of a word must be ascertained in the context of achieving particular objectives, and the words associated with it may indicate that the true meaning of the series is to convey a common idea. The language may reasonably be interpreted to impose the requirement on any discrete, but integrated, operation which pollutes. This gives meaning to all of the terms — a single building, not part of a larger operation, would be covered if it emits more than 100 tons of pollution, as would any facility, structure, or installation. Indeed, the language itself implies a “bubble concept” of sorts: each enumerated item would seem to be treated as if it were encased in a bubble. While respondents insist that each of these terms must be given a discrete meaning, they also argue that § 111(a)(3) defines “source” as that term is used in § 302(j). The latter section, however, equates a source with a facility, whereas the former defines “source” as a facility, among other items.

We are not persuaded that parsing of general terms in the text of the statute will reveal an actual intent of Congress.34 *862We know full well that this language is not dispositive; the terms are overlapping and the language is not precisely-directed to the question of the applicability of a given term in the context of a larger operation. To the extent any congressional “intent” can be discerned from this language, it would appear that the listing of overlapping, illustrative terms was intended to enlarge, rather than to confine, the scope of the agency’s power to regulate particular sources in order to effectuate the policies of the Act.

Legislative History

In addition, respondents argue that the legislative history and policies of the Act foreclose the plantwide definition, and that the EPA’s interpretation is not entitled to deference because it represents a sharp break with prior interpretations of the Act.

Based on our examination of the legislative history, we agree with the Court of Appeals that it is unilluminating. The general remarks pointed to by respondents “were obviously not made with this narrow issue in mind and they cannot be said to demonstrate a Congressional desire . . . .” Jewell Ridge Coal Corp. v. Mine Workers, 325 U. S. 161, 168-169 (1945). Respondents’ argument based on the legislative history relies heavily on Senator Muskie’s observation that a new source is subject to the LAER requirement.35 But the full statement is ambiguous and like the text of § 173 itself, this comment does not tell us what a new source is, much less that it is to have an inflexible definition. We find that the legislative history as a whole is silent on the precise issue before us. It is, however, consistent with the view that the EPA should have broad discretion in implementing the policies of the 1977 Amendments.

*863More importantly, that history plainly identifies the policy concerns that motivated the enactment; the plantwide definition is fully consistent with one of those concerns — the allowance of reasonable economic growth — and, whether or not we believe it most effectively implements the other, we must recognize that the EPA has advanced a reasonable explanation for its conclusion that the regulations serve the environmental objectives as well. See supra, at 857-859, and n. 29; see also supra, at 855, n. 27. Indeed, its reasoning is supported by the public record developed in the rulemaking process,36 as well as by certain private studies.37

Our review of the EPA’s varying interpretations of the word “source” — both before and after the 1977 Amendments — convinces us that the agency primarily responsible for administering this important legislation has consistently interpreted it flexibly — not in a sterile textual vacuum, but in the context of implementing policy decisions in a technical and complex arena. The fact that the agency has from time to time changed its interpretation of the term “source” does not, as respondents argue, lead us to conclude that no deference should be accorded the agency’s interpretation of the statute. An initial agency interpretation is not instantly carved in stone. On the contrary, the agency, to engage in informed rulemaking, must consider varying interpretations *864and the wisdom of its policy on a continuing basis. Moreover, the fact that the agency has adopted different definitions in different contexts adds force to the argument that the definition itself is flexible, particularly since Congress has never indicated any disapproval of a flexible reading of the statute.

Significantly, it was not the agency in 1980, but rather the Court of Appeals that read the statute inflexibly to command a plantwide definition for programs designed to maintain clean air and to forbid such a definition for programs designed to improve air quality. The distinction the court drew may well be a sensible one, but our labored review of the problem has surely disclosed that it is not a distinction that Congress ever articulated itself, or one that the EPA found in the statute before the courts began to review the legislative work product. We conclude that it was the Court of Appeals, rather than Congress or any of the decision-makers who are authorized by Congress to administer this legislation, that was primarily responsible for the 1980 position taken by the agency.

Policy

The arguments over policy that are advanced in the parties’ briefs create the impression that respondents are now waging in a judicial forum a specific policy battle which they ultimately lost in the agency and in the 32 jurisdictions opting for the “bubble concept,” but one which was never waged in the Congress. Such policy arguments are more properly addressed to legislators or administrators, not to judges.38

*865In these cases the Administrator’s interpretation represents a reasonable accommodation of manifestly competing interests and is entitled to deference: the regulatory scheme is technical and complex,39 the agency considered the matter in a detailed and reasoned fashion,40 and the decision involves reconciling conflicting policies.41 Congress intended to accommodate both interests, but did not do so itself on the level of specificity presented by these cases. Perhaps that body consciously desired the Administrator to strike the balance at this level, thinking that those with great expertise and charged with responsibility for administering the provision would be in a better position to do so; perhaps it simply did not consider the question at this level; and perhaps Congress was unable to forge a coalition on either side of the question, and those on each side decided to take their chances with the scheme devised by the agency. For judicial purposes, it matters not which of these things occurred.

Judges are not experts in the field, and are not part of either political branch of the Government. Courts must, in some cases, reconcile competing political interests, but not on the basis of the judges’ personal policy preferences. In contrast, an agency to which Congress has delegated policymak-ing responsibilities may, within the limits of that delegation, properly rely upon the incumbent administration’s views of wise policy to inform its judgments. While agencies are not directly accountable to the people, the Chief Executive is, and it is entirely appropriate for this political branch of the Government to make such policy choices — resolving the competing interests which Congress itself either inadvertently did not resolve, or intentionally left to be resolved by the *866agency charged with the administration of the statute in light of everyday realities.

When a challenge to an agency construction of a statutory provision, fairly conceptualized, really centers on the wisdom of the agency’s policy, rather than whether it is a reasonable choice within a gap left open by Congress, the challenge must fail. In such a case, federal judges — who have no constituency — have a duty to respect legitimate policy choices made by those who do. The responsibilities for assessing the wisdom of such policy choices and resolving the struggle between competing views of the public interest are not judicial ones: “Our Constitution vests such responsibilities in the political branches.” TVA v. Hill, 437 U. S. 153, 195 (1978).

We hold that the EPA’s definition of the term “source” is a permissible construction of the statute which seeks to accommodate progress in reducing air pollution with economic growth. “The Regulations which the Adminstrator has adopted provide what the agency could allowably view as . .. [an] effective reconciliation of these twofold ends .. . .” United States v. Shimer, 367 U. S., at 383.

The judgment of the Court of Appeals is reversed.

It is so ordered.

Justice Marshall and Justice Rehnquist took no part in the consideration or decision of these cases.

Justice O’Connor took no part in the decision of these cases.

Opinion Per Curiam.

PER CURIAM:

This is a petition for review of an order of the Environmental Protection Agency implementing the 1990 amendments to § 112 of the Clean Air Act. Petitioners are General Electric Company and four trade associations: (1) National Mining Association, which represents companies that produce metal, coal, and minerals, and that manufacture mining equipment; (2) American Forest and Paper Association, which represents companies that make pulp, paper, paperboard, and solid wood; (3) Chemical Manufacturers Association, which represents companies that manufacture industrial chemicals; and (4) American Petroleum Institute, which represents companies engaged in the petroleum industry. We deny the petition for review with respect to the issues raised by General Electric, National Mining Association, and American Forest and Paper Association, but grant it with respect to Chemical Manufacturers Association and American Petroleum Institute’s challenge.

I

In 1990, as part of its comprehensive overhaul of the Clean Air Act, Pub.L. No. 91-604, *135384 Stat. 1676 (1970), Congress revised § 112 of the Act, which regulates emissions of hazardous air pollutants. Pub.L. No. 101-549, 104 Stat. 2399, 2531-84 (1990). Dissatisfied with EPA’s health-based regulation of hazardous air pollutants under the 1970 program,1 S.Rep. No. 228, 101st Cong., 1st Sess. 128 (1989), U.S.Code Cong. & Admin.News 1990, 3385, Congress replaced this approach with a detailed, technology-based regulatory scheme. The 1990 amendments to § 112 establish an initial list, which EPA may periodically revise, of 189 hazardous air pollutants. 42 U.S.C. § 7412(b)(l)-(3). EPA must publish a list of “categories and subcategories” of “major sources” and certain “area sources” that emit these pollutants. 42 U.S.C. § 7412(e). For each listed “category or subcategory of major sources and area sources” of hazardous air pollutants, § 112(d) of the Act directs EPA to promulgate emission standards. 42 U.S.C. § 7412(d)(1).

Under the Act, “major sources” of hazardous air pollutants are potentially subject to stricter regulatory control than are “area sources.”2 For example, major sources must comply with technology-based emission standards requiring the maximum degree of reduction in emissions EPA deems achievable, often referred to as “maximum achievable control technology” or MACT standards.3 42 U.S.C. § 7412(d)(1) — (2). In order to obtain an operating permit under title Y of the Act, §§ 501-507, major sources must comply with extensive monitoring, reporting and record-keeping requirements. 42 U.S.C. §§ 7661-76611 Further, § 112(g) generally conditions the modification, construction or reconstruction of a major source on the source’s meeting MACT emission limitations. 42 U.S.C. § 7412(g).

“Area sources” of hazardous air pollutants are not necessarily subject to such stringent regulation. EPA need not list all “categories and subcategories” of area sources, 42 U.S.C. § 7412(e)(3),4 and it does not have to establish emission standards for unlisted area sources, 42 U.S.C. § 7412(d)(1). For listed area sources, EPA may choose to promulgate emission standards requiring only “generally available control technologies or management practices.” 42 U.S.C. § 7412(d)(5). These standards can be less rigorous than those required for major sources under 42 U.S.C. *1354§ 7412(d)(1). S.Rep. No. 228, supra, at 172. Area sources are not subject to title V permitting requirements, or to § 112(g)’s restrictions on modification, construction and reconstruction of their facilities.

In July 1992, pursuant to § 112(c)(1), EPA published an initial list of categories of sources that emit hazardous air pollutants, 57 Fed.Reg. 31,576 (1992), and almost seventeen months later, it published a schedule for promulgation of emission standards for these listed source categories, as required by § 112(e), 58 Fed.Reg. 63,941 (1993). In August 1993, in order to “eliminate the need to repeat general information and requirements within each [emission] standard,” EPA proposed a rule codifying the “procedures and criteria needed to implement” emission standards for hazardous air pollutants. 58 Fed. Reg. 42,760, 42,760 (1993). It promulgated a final rule, which is the subject of this dispute, adopting these general provisions on March 16, 1994. 59 Fed.Reg. 12,408 (1994).

Among other things, the general provisions rule implements § 112(a)(l)’s definition of “major source.” The rule defines “major source” in terms nearly identical to those in § 112(a)(1) of the Clean Air Act:

Major source means any stationary source or group of stationary sources located within a contiguous area and under common control that emits or has the potential to emit considering controls, in the aggregate, 10 tons per year or more of any hazardous air pollutant or 25 tons per year or more of any combination of hazardous air pollutants, unless the Administrator establishes a lesser quantity, or in the case of radionuclides, different criteria from those specified in this sentence.

59 Fed.Reg. at 12,433-34 (to be codified at 40 C.F.R. § 63.2). A “stationary source” is “any building, structure, facility or installation which emits or may emit any air pollutant.” Id. An “area source [is] any stationary source ... that is not a major source.” Id. The preambles to the proposed and final rules, and other definitions adopted in the final rule explain in greater detail how EPA plans to identify major sources.

Petitioners challenge three aspects of EPA’s implementation of the definition of “major source.” First, National Mining Association and American Forest and Paper Association (collectively referred to as “National Mining Association”) and General Electric question EPA’s requiring the aggregation of all hazardous air emissions within a plant site — instead of only those emissions from equipment in similar industrial categories — in a § 112 major source determination. Second, National Mining Association challenges EPA’s requiring the inclusion of “fugitive emissions” in a source’s aggregate emissions in determining whether the source is major. Third, Chemical Manufacturers Association and American Petroleum Institute (collectively referred to as “Chemical Manufacturers Association”) contend that EPA overstepped its regulatory authority by permitting a source to reduce its “potential to emit” only with “federally enforceable” emission controls and limitations.

II

EPA promulgated the rule challenged here in accordance with the special rulemaking provisions of 42 U.S.C. § 7607(d). EPA rules that are “arbitrary, capricious, an abuse of discretion, or otherwise not in accordance with law” or in excess of EPA’s “statutory jurisdiction, authority, or limitations” must be set aside. 42 U.S.C. § 7607(d)(9). With respect to alleged procedural errors in EPA’s promulgation of a rule, these will invalidate the rule only “if the errors were so serious and related to matters of such central relevance to the rule that there is a substantial likelihood that the rule would have been significantly changed if such errors had not been made.” 42 U.S.C. § 7607(d)(8).

A

General Electric and National Mining Association have similar arguments against the final rule’s implementation of § 112(a)(1). Both maintain that EPA may not, in determining whether a site is a major source, include emissions from all facilities on a contiguous plant site under common control. These petitioners assert that, for purposes of major source determinations, EPA may aggregate emissions from different facilities on *1355a contiguous plant site under eon. non control only when the facilities fall within a similar industrial classification. General Electric says EPA must aggregate emissions on a “source category” basis;5 National Mining Association contends that EPA may combine emissions only if the emitting facilities fall within the same two-digit Standard Industrial Classification (SIC) Code.6

In the preamble to the final rule,7 EPA made clear that in determining whether a source is major, emissions from all sources of hazardous air pollutants within a plant site must be aggregated, so long as the sources are geographically adjacent and under common control. 59 Fed.Reg. at 12,412. As a result, if the total annual emissions of hazardous air pollutants from a plant site exceed the designated thresholds, each source emitting pollutants at the site must comply with the stricter MACT emission standards applicable to sources under § 112(d)(2), and with other requirements applicable to major sources.

Petitioners read § 112(a)(1) more restrictively. In their view, EPA’s approach will impermissibly regulate “minor facilities” that happen to be located at an industrial site with annual emissions of hazardous air pollutants that, in the aggregate, exceed the major source thresholds. See Brief for General Electric at 19. They contend that EPA may require aggregation of emissions from sources only if those sources fall within a single source category — General Electric’s argument, or the same two-digit SIC Code— National Mining Association’s contention. It follows, according to petitioners, that a source must comply with regulatory requirements applicable to major sources only if it belongs to some group of sources at an industrial site emitting, in the aggregate, more than the major source threshold. Under petitioners’ theories, it is possible that only some of a site’s sources would have to comply with the regulatory requirements applicable to major sources, including the stricter emission limitations of § 112(d)(2). Other sources of hazardous air pollutants would be regulated as area sources, possibly subject to less stringent emission standards or to none at all. 42 U.S.C. § 7412(e)(5).

EPA rejected petitioners’ methods of implementing “major source.” With respect to General Electric’s source category definition, EPA acknowledged that “[m]ore than one source category on the EPA’s source category list may be represented within a plant that is a major source” of hazardous air pollutants, as is the case for a large chemical manufacturing complex. 59 Fed.Reg. at 12,-411; see also 57 Fed.Reg. at 31,578 (“a large plant ... would clearly be a ‘major source,’ but would also comprise multiple source categories”). Congress intended, according to EPA, “that all portions of a major source be subject to MACT [emission standards] regardless of the number of source categories into which the facility is divided.” 59 Fed. Reg. at 12,411. “Thus, the EPA will set one or more MACT standards for a major source, and sources within that major source will be covered by the standard(s), regardless of whether, when standing alone, each one of *1356those regulated sources would be major.” Id. EPA also rejected the SIC Code approach to implementing “major source,” advanced here by National Mining Association. Because § 112(a)(1) does not refer to SIC Codes, EPA reasoned that Congress intended major sources of hazardous air pollutants to “encompass entire contiguous ... plant sites without being subdivided according to industrial classifications.” 59 Fed.Reg. at 12,412. A separation of emission sources by SIC Codes “would be an artificial division of sources that, in reality, all contribute to public exposure around a plant site.” Id.

If § 112(a)(1) is viewed in isolation, EPA’s reading of the provision is not simply consistent with the provision; it is nearly compelled by the statutory language. Section 112(a)(1) states that a “group of stationary sources” need meet only three conditions to be termed a “major source”: (1) sources within the group must be “located within a contiguous area”; (2) they must be “under common control”; and (3) in the aggregate, they must emit or, considering controls, have the potential to emit 10 or more tons per year of a single hazardous air pollutant or 25 or more tons per year of any combination of hazardous air pollutants. Section 112(a)(1) says nothing about combining emissions only from sources within the same source categories or SIC Codes. In this respect, EPA’s definition of “major source,” set forth in the preamble to the final rule, is faithful to the language of § 112(a)(1).

Petitioners ask us to look beyond the language of § 112(a)(1). In the first of several loosely connected arguments, General Electric recites fragments from § 112’s other provisions, including: (1) § 112(e)(1), which directs EPA to publish “a list of all categories and subcategories of major sources and area sources”; (2) § 112(d)(1), which directs EPA to establish emission standards “for each category or subcategory of major sources and area sources”; and (3) § 112(j)(2), which describes what an operator of a “major source in [a] category” must do if EPA does not promulgate an emission standard for that “category of major sources.” From these portions of § 112, General Electric leaps to the conclusion that “major source must be defined “with reference to’ (and cannot be broader than) the source category defined by EPA for § 112 regulation.” Brief for General Electric at 16.

General Electric’s logic is hard to grasp. Rather than supporting General Electric, the provisions the company invokes, read in full and in context, tend to support EPA’s implementation of “major source” without reference to source categories. Section 112 directs EPA to perform certain tasks on a category-wide basis — it is to identify categories of major and area sources of hazardous air pollutants (§ 112(c)(1)), and it must promulgate category-wide emission standards for these sources (§ 112(d)(2)). It by no means follows that because the statute in several provisions uses the terms “major source” and “category” in the same sentence — which is all General Electric’s argument amounts to — EPA must read a source category restriction into § 112(a)(l)’s definition of “major source.” Nor does § 112(c) somehow prohibit EPA from applying § 112(d)’s MACT emission limitations “to minor sources in a listed category of major sources without complying with the statutory requirements for listing area sources.” See Brief for General Electric at 18. Section 112(c) simply requires the listing of all major sources and those area sources presenting adverse health or environmental effects. 42 U.S.C. § 7412(c)(1), (3). Neither § 112(c) nor § 112(d) says anything about EPA’s including “minor sources” in a “listed category of major sources.”8

Taking its argument one step further, General Electric contends that EPA’s definition *1357of “major source” will lead to “anomalous and unreasonable results” when other parts of §112 are implemented. Brief for General Electric at 19. General Electric sees a tension between the final rule’s definition of “major source” and § 112(g), which describes requirements for constructing, reconstructing or modifying a major source. 42 U.S.C. § 7412(g). The idea is that under § 112(g), an operator of a small emissions unit at a large facility might have to install MACT in order to modify or reconstruct the unit, even though the unit would not be subject to a MACT standard under § 112(d). We agree with EPA that General Electric’s argument reflects a misreading of § 112, which draws no distinction between “major source” for purposes of § 112(d) and § 112(g). If a small emissions unit is a “major source” because it is located at a plant that emits or, considering controls, has the potential to emit 10 or more tons per year of a single hazardous air pollutant or 25 or more tons per year of any combination of hazardous air pollutants, it is subject to all the regulatory requirements imposed on major sources, including those of § 112(d) and § 112(g).9

In addition to alleging inconsistencies between EPA’s definition of “major source” and § 112’s other provisions, General Electric insists that EPA’s implementation of § 112(a)(1) is at odds with other aspects of the Clean Air Act. It points to § 112(a)(3)’s provision that “stationary source” has the same meaning as it does in § 111 of the Act, 42 U.S.C. § 7411, which deals with performance standards applicable to new sources. In Alabama Power Co. v. Costle, 636 F.2d 323, 395-96 (D.C.Cir.1979), General Electric notes, this court rejected EPA’s definition of “stationary source” as used in § 111(a)(3) as a “combination” of “buildings, structures, facilities, or installations.”10 To be sure, Alabama Power struck down EPA’s defining “source” for purposes of its preventing significant deterioration in air quality program (established pursuant to 42 U.S.C. §§ 7470-7479)11 as any “structure, building, facility equipment, installation or operation (or combination thereof).” 636 F.2d at 394, 395-96 (italics added). But that was because EPA had unlawfully expanded § 111(a)(3), which defines “stationary source” as “any building, structure, facility or installation which emits or may emit any air pollutant,” without reference to combinations of these things. 636 F.2d at 395. That is not the case here. Section 112(a)(1) expressly provides that a “major source” is “any stationary source or group of stationary sources ” with emissions exceeding certain limits. 42 U.S.C. § 7412(a)(1) (italics added). Indeed, one could infer from § 112(a)(1) a congressional intent, in the context of hazardous air pollution regulation, to override Alabama Power.12

We also reject General Electric’s final argument that EPA’s implementation of § 112(a)(1) was procedurally flawed. EPA sufficiently addressed the effect of its definition of “major source” on various aspects of the hazardous air pollution program, and it was explicit enough about binding effect of the final rule’s definition on future actions. In both the proposed and final rules, EPA *1358extensively discussed the implications of the general provisions, including its definition of “major source,” on other aspects of § 112. See 59 Fed.Reg. at 12,414-18; 58 Fed.Reg. at 42,764-68. The agency further explained that it intended the general provisions to be “the minimum generic requirements necessary for implementation” of emission standards for hazardous air pollutants. 59 Fed. Reg. at 12,415. To the extent EPA finds it appropriate to override specific aspects of the final rule in future rulemakings for specific source categories, “EPA will describe in the [new] subpart exactly which requirements of the General Provisions are applicable to the specific source category and which have been overridden.” Id. at 12,408-09. Even if General Electric were correct about EPA’s alleged procedural defects, the company has not shown why these supposed mistakes were so serious that, had they not been made, there is a “substantial likelihood that the rule would have been significantly changed.” See 42 U.S.C. § 7607(d)(8).

National Mining Association takes a somewhat different tack in its challenge to EPA’s definition of “major source.” For serious, severe and extreme ozone nonattainment areas, § 182(c)-(e) (42 U.S.C. § 7511a(e)-(e)), and for the title V permitting program, § 501(2) (42 U.S.C. § 7661(2)), the Act defines the terms “major source” or “major stationary source” in language very similar to that of § 112(a)(1). See, e.g., 42 U.S.C. § 7511a(c) (defining “major source” as “any stationary source or group of sources located within a contiguous area and under common control” with emissions exceeding 50 tons per year of volatile organic compounds). As EPA acknowledges, to some extent it uses a SIC Code approach to defining “major source” in these programs. See, e.g., 40 C.F.R. § 70.2 (emissions to be aggregated by SIC Code for title V major source determination).13 National Mining Association reasons that EPA is required to use the same approach in defining “major source” pursuant to § 112(a)(1).

This argument warrants little discussion. As EPA explained in the preamble to the final rule, “because of the different objectives of section 112 and title V ... and because section 112 contains its own definition,” it would define “major source” for purposes of § 112 without reference to SIC Codes. 59 Fed.Reg. at 12,412. The explanation is reasonable. Different programs have different objectives and structures. EPA is not bound to any one definition of “major source.” See Alabama Power, 636 F.2d at 397-98 (“EPA has latitude to adopt definitions of ... ‘source’ that are different ... from those [used in other programs].”); of. Mobil Oil Corp. v. EPA, 871 F.2d 149, 153 (D.C.Cir.1989) (“This court has previously upheld the agency’s decision to employ different definitions of the term ‘facility’ in construing different portions of RCRA.”). The § 112 and § 182 (42 U.S.C. § 7511a) programs are plainly distinguishable. Section 112 is directed at limiting nationwide emissions of hazardous air pollutants; § 182 addresses emissions of volatile organic compounds in serious, severe and extreme ozone nonattainment areas. Apart from this very obvious difference, in § 182 the term “major source” is used mainly to determine the sort of control technology that must be installed by an operator wishing to modify a facility in a way that increases emissions of volatile organic compounds. See, e.g., 42 U.S.C. § 7511a(e)(7). Whether a source is major for purposes of § 112 governs not only modifications, § 112(g), but also emission limitations applicable to new and existing sources of hazardous air pollutants, § 112(d). The title V program also differs fundamentally from § 112. It is essentially procedural — other provisions in the Act supply the substantive requirements applicable to sources subject to the permitting program.

In a second argument, National Mining Association contends that EPA’s definition of “major source” is inconsistent with the legislative history of the 1990 amendments to the Clean Air Act. It bases this conclusion on an excerpt from a House Report discussing the “major source” definitional language added *1359to the ozone nonattainment provision in § 182 of the Act, 42 U.S.C. § 7511a:

The definition of “major source” [in the ozone nonattainment area] and elsewhere in the bill uses the term “group of sources located within a contiguous area and under common control.” The Committee understands this to mean a group of sources with a common industrial grouping, i.e., the same two-digit SIC code. It is the approach followed by EPA as a result of the Alabama Power litigation. It avoids the possibility that dissimilar sources, like a power plant and an adjacent coal mine, will be considered as the same “source” because of common ownership.

H.R.Rep. No. 490(1), 101st Cong., 2d Sess. 236-37 (1990) (italics added). Seizing upon the italicized language, National Mining Association urges that “Congress intended that a source-based limitation, such as the SIC limitation, apply to all of the definitions” of “major source.” Brief for National Mining Association at 21.

We agree with EPA that the legislative history of the Clean Air Act does not compel a conclusion that Congress intended to limit “major source” in the way National Mining Association suggests. The section of the Report that specifically discusses § 112(a)(l)’s definition of “major source” contradicts the Association’s interpretation of the provision:

For purposes of the definition [of “major source”], all emissions of listed pollutants are counted from a group of sources within a plant boundary (contiguous property under common ownership). This is to assure that emissions from the facility as a whole are adequately controlled.

H.R.Rep. No. 490(1), supra, at 324 (italics added). The Senate Report contains similar language:14

The definition of “major source” [in § 112] also includes provisions to assure that stationary sources which would otherwise be subject to the emissions standards, are not excluded from control requirements as the result of arbitrary subdivision or description of the source. A stationary source potentially subject to an emissions standard because it emits a listed air pollutant is to be defined to include “all emission points and units of such source located within a contiguous area and under common control”. This language will prevent a facility from avoiding control by subdividing its operations into separate emissions points, product lines or units at one site with individual emissions rates less than the 10 or 25 ton per year thresholds.

S.Rep. No. 228, supra, at 151, U.S.Code Cong. & Admin.News 1990, at 3536 (italics added). These explicit references to § 112(a)(l)’s definition of “major source” support EPA’s view. At most, the legislative history leaves unresolved whether the agency must define “major source” with reference to SIC Codes. In such circumstances, “it is enough that the Agency’s construction is reasonable.” Natural Resources Defense Council v. EPA, 22 F.3d 1125, 1141 (D.C.Cir.1994) (citing Ohio v. EPA, 997 F.2d 1520, 1527 (D.C.Cir.1993)). EPA’s interpretation of “major source” satisfies that standard.15

B

National Mining Association also thinks EPA erred in deciding to count “fugitive emissions” of hazardous air pollutants in determining whether a “source” is a “major source,” without first conducting a rulemaking pursuant to § 302(j), 42 U.S.C. § 7602(j). “Fugitive emissions” are defined in the final rule as:

those emissions from a stationary source that could not reasonably pass through a stack, chimney, vent or other functionally *1360equivalent opening. Under section 112 of the Act, all fugitive emissions are to be considered in determining whether a stationary source is a major source.

59 Fed.Reg. at 12,433 (to be codified at 40 C.F.R. § 63.2).16

Section 302(j) of the Act, as interpreted in Alabama Power v. Costle, 636 F.2d at 369-70, requires EPA to conduct a separate rule-making to achieve this result, so the Association contends. This provision, fully quoted in the margin,17 states that, “[ejxcept as otherwise expressly provided,” a “major stationary source” or “major emitting facility” is any stationary source of air pollutants that “directly emits, or has the potential to emit” at least 100 tons per year of any air pollutant, including “any major ... source of fugitive emissions ... as determined by rule by the Administrator.” Alabama Power held that EPA could not, without rulemaking, include fugitive emissions of air pollutants in a facility’s aggregate emissions for purposes of determining whether the facility was a “major emitting facility” within § 169(1), 42 U.S.C. § 7479. 636 F.2d at 368-70. For the prevention of significant deterioration in air quality program (42 U.S.C. §§ 7470-7479), § 169(1) defines “major emitting facility” as any of 28 categories of sources that emit 100 or more tons per year of any air pollutant. 42 U.S.C. § 7479(1). For “any other source,” the threshold is 250 or more tons per year. Id. Although § 169(1) did “expressly make a substantial modification in the 302(j) definition of ‘major,’ ” it had “no ‘express’ provision modifying section 302(j)’s ‘rule’ requirement as to fugitive emissions.” Alabama Power, 636 F.2d at 370. “Therefore under section 169(1) controlled in this respect by section 302(j), the calculation of the threshold quantity emissions may include fugitive emissions only as determined by rule by the Administrator.” 18 636 F.2d at 370.

Finding this case indistinguishable from Alabama Power, National Mining Association spins out the following argument: § 112(a)(1) defines “major source” in terms of a “stationary source or group of stationary sources”; a § 112 major source is thus, “by definition, a ‘major stationary source,’ ” subject to the requirements of § 302(j); § 112(a)(1) does not expressly exempt a major source from § 302(j)’s fugitive emissions rulemaking requirement; therefore, EPA may not require a source to include fugitive emissions of hazardous air pollutants in the source’s total emissions without a § 302(j) rulemaking.19

The Association’s argument is not very persuasive. Alabama Power was decided in the pre-Chevron age. Moreover, there is a notable difference between § 302(j) and § 112(a)(1). Section 302(j) speaks of sources that “directly” emit air pollutants, on the one hand, and fugitive emissions on the other, thus suggesting that emissions of the fugitive variety are not direct. By contrast, § 112(a)(1) does not contain the modifier “di*1361reetly,” and it does not mention fugitive emissions as a separate category of emissions. Furthermore, one cannot say that § 302(j) supplies “quantitative terms” for § 112(a)(l)’s definition of “major source,” as it did for § 169(1), one of the provisions at issue in Alabama Power, 636 F.2d at 369. When it comes to hazardous air pollutants, the quantities — 10 tons of any one kind per year or 25 tons of a combination — are specified in § 112(a)(1), not § 302(j), which has a 100-ton threshold.

While the Association’s argument is thin, EPA’s counterargument is hardly overwhelming. EPA thinks its best point is that § 112(a)(1) defines “major source” whereas § 302(j) defines “major stationary source” and “major emitting facility”; since the phrases are different so must be the meanings with respect to counting fugitive emissions. The problem with this argument is that at a critical juncture § 302(j) also uses the phrase “major source” — a “major stationary source” may include a “major ... source of fugitive emissions” if EPA so decides in a rulemaking. And as the Association notes, other provisions of the Act unrelated to § 112 or § 302(j) refer to “major source” and “major stationary source” interchangeably, see 42 U.S.C. §§ 7426(a)(1)-(2), 7511a, 7513a. If EPA’s point related to differences between major stationary sources and major mobile sources, the omission of “stationary” in § 112(a)(1) and its inclusion in § 302(j) might be significant, but that of course is not EPA’s point.

EPA fares better when it tells us that title V of the Act explicitly draws a distinction between the nomenclature of § 112 and that of § 302, defining “major source” for permitting purposes as either a “major source as defined in section 7412 [§ 112]” or a “major stationary source as defined in section 7602 [§ 302] ... or part D of subchapter I [nonattainment program].” 42 U.S.C. § 7661(2). And on EPA’s side is the Senate committee report stating that the definition of “major source” in § 112 “will only apply in the context of this section and should not be confused with other meanings of the term ‘major source’ in parts C (prevention of significant deterioration) or D (non-attainment) of the Act.” S.Rep. No. 228, supra, at 150-51, U.S.Code Cong. & Admin.News 1990, at 3535-3536.

We conclude that EPA may require the inclusion of fugitive emissions in a site’s aggregate emissions without conducting any special rulemaking, even if “major source” and “major stationary source” mean the same thing. Section 112(a)(1) expressly provides that a “major source” is any stationary source or group of stationary sources “located within a contiguous area and under common control” and emitting more than 10 tons per year of a single hazardous air pollutant or 25 tons per year of such pollutants combined. An emission may be fugitive, but it is still an emission from a stationary source. And so the italicized language certainly may be read as EPA reads it — that all emissions are to be counted in determining whether a source is major, subject only to the qualification that they emanate from a contiguous site under common control. So read, § 112(a)(1) satisfies § 302(j)’s “[e]xcept as otherwise expressly provided” clause such that fugitive emissions may be counted in a source’s aggregate emissions without a special rulemaking.

C

As noted above, in determining whether a source is to be categorized as a “major source” of emissions (or by default an “area” source), EPA was directed by Congress to calculate the amount of hazardous air pollutants a stationary source “emits or has the potential to emit considering controls.” Clean Air Act § 112(a)(1), 42 U.S.C. § 7412(a)(1) (emphasis added). In its final rule, EPA defined a source’s “potential to emit” as its “maximum capacity ... to emit a pollutant under its physical and operational design.” 59 Fed.Reg. at 12,434. To comply with the statutory directive to “consider[] controls” while determining emissions capacities, the rule also provides:

Any physical or operational limitation on the capacity of the stationary source to emit a pollutant, including air pollution control equipment and restrictions on hours of operation or on the type or amount or material combusted, stored, or *1362processed, shall be treated as part of its design if the limitation or the effect it would have on emissions is federally enforceable.

Id. (emphasis added). Under the rule, a control is deemed to be “federally enforceable” if it is “enforceable by the Administrator and citizens under the Act or ... under other statutes administered by the Administrator.” Id. at 12,433.

Petitioner Chemical Manufacturers Association argues that this restrictive definition— which disregards emissions limitations imposed by state or local regulations not deemed “federally enforceable” — is contrary to the language of § 112(a)(1) of the Act. The government contends that since the word “controls” is not defined in the statute, it was open to EPA under Chevron to define the term, and it has done so reasonably. See Chevron U.S.A. Inc. v. Natural Resources Defense Council, Inc., 467 U.S. 837, 843, 104 S.Ct. 2778, 2782, 81 L.Ed.2d 694 (1984). According to petitioners, even if Chevron Step II is to be reached — because the statute does not reveal a specific congressional intent — we should conclude that EPA’s construction of “controls” is impermissible.

It is common ground that Congress meant the word “controls” to refer to governmental regulations and not, for instance, operational restrictions that an owner might voluntarily adopt. (We note, however, that the word could be read that broadly, which certainly supports the government’s position that the term is not clear on its face.) Petitioners further conceded at oral argument — quite properly, we believe — that Congress intended the term to stand for effective controls. EPA clearly is not obliged to take into account controls that are only chimeras and do not really restrain an operator from emitting pollution. Nevertheless, petitioners claim that EPA has imposed the federal enforceability requirement in pursuit of policy objectives unrelated to concerns about the effectiveness of controls imposed at the state and local level. EPA is accused of interpreting the statute so as to pressure states — through the ministrations of sources eager to have local controls counted in determining their capacity to emit under § 112 — to seek EPA approval of state emissions policies. This objective, petitioners claim, is no part of § 112’s requirement that controls be considered in determining whether a facility qualifies as a major source. It is an impermissible interpretation since it subordinates the effectiveness of controls to other considerations not approved by Congress.

Although it is the regulations implementing the 1990 amendments to the Clean Air Act which are directly before us, this dispute had its genesis at least a decade earlier. Following the passage of the Clean Air Act Amendments of 1977, the agency took the position that the phrase “potential to emit” as used in the definition of “major emitting facilities” excluded even emissions-reducing equipment, such as scrubbers, filters, and other technologies. See 40 C.F.R. §§ 51.24(b)(3), 52.21(b)(3) (1978). We rejected that position in Alabama Power. See 636 F.2d at 353-55. In the wake of that case, EPA proposed a new definition of “potential to emit” that would have taken into account air pollution control equipment, but not operational restraints. See 44 Fed.Reg. 51,924 (1979). The final regulations issued in 1980, however, adopted the position that capacity calculations could factor in operational restraints — but only if they were “federally enforceable.” See 45 Fed.Reg. 52,676, 52,746 (1980). The regulations defined as “federally enforceable” those emissions restrictions that were “enforceable by the Administrator.” Id. at 52,737. The requirement of federal enforceability was, EPA explained, “necessary, as a practical matter, to ensure that sources will perform the proper operation and maintenance for the control equipment.” Id. at 52,688.

The 1980 rule was challenged in this court, but in a February 1982 settlement, EPA agreed to amend its position on federal enforceability. The proposal that followed would have taken into account emission limits “enforceable under federal, state or local law and discoverable by the Administrator and any other person.” 48 Fed.Reg. 38,742, 38,-748, 38,755 (1983). But by the time the final rule was issued, in 1989, the agency had apparently decided to abandon the terms of the settlement. The final regulations revert*1363ed to the former position of requiring federal enforceability as the sine qua non for crediting operational restraints. “Federal enforceability” was still defined to reach only those limitations “enforceable by the Administrator,” but this term now included state constraints imposed under federally approved plans. See 54 Fed.Reg. 27,274, 27,285-86 (1989). New litigation followed but the cases were stayed (in our court) in anticipation of the 1990 amendments.

Congress thus acted in 1990 against a backdrop of over a decade of skirmishing between the agency and affected companies, during which the issue of whether and to what extent state and local controls were to be credited in calculating a source’s “potential to emit” was very much in the forefront. In drafting § 112 Congress specifically directed EPA to consider controls in determining which producers should be classified as “major sources,” but conspicuously did not limit controls to those that are federally enforceable. The government maintains that since Congress did not specify what kind of controls would qualify, EPA was free to answer that question. It permissibly did so by once again requiring that they had to be “federally enforceable,” a term which, in EPA’s final manifestation of the concept, applies to “all limitations and conditions that are enforceable by the Administrator and citizens under the Act or that are enforceable under other statutes administered by the Administrator.” 54 Fed.Reg. at 12,433.

As we have noted, it is certainly permissible for EPA to have refused to take into account ineffective controls (indeed, it is likely that a contrary interpretation would be impermissible). But is it also open to EPA under the statute to refuse to consider controls on grounds other than their lack of effectiveness? To qualify as “federally enforceable,” (as best we can determine) controls are required, in addition to being effective as a practical matter, to have been approved by EPA and integrated into the state implementation plan, or SIP, drawn up by each state to enforce substantive restrictions under the Clean Air Act and submitted to the Administrator for approval under § 110, 42 U.S.C. § 7410. Once included within the SIP, a state control becomes enforceable not only by the state which is its primary regulating authority, but also by the Administrator under § 113 of Act, 42 U.S.C. § 7413, and, in certain settings, by private citizens, who can bring suit for noncompliance with federal pollution control programs under § 304, 42 U.S.C. § 7604.

EPA has identified several state and local regulatory approaches through which states can impose restraints and have them deemed “federally enforceable.” Constraints imposed upon a source under a state operating permit, for example, will be deemed “federally enforceable” if the state program has been approved as a “federally enforceable state operating permit program,” or FESOPP, by EPA. A state permitting program cannot stand alone; it must be incorporated into the SIP, must impose upon sources a legal obligation to observe the permit constraints, must be enforceable as a practical matter— i.e., must be “effective” — must not be inconsistent with other requirements under the SIP or federal law, and must be issued pursuant to a public hearing process. See 54 Fed.Reg. 27,274,27,281-82 (1989). Other approaches are available as well. General, as opposed to source-specific, permits can also be issued under a FESOPP, or under a state general permitting program similarly approved for inclusion within the SIP. And a state can impose constraints by general prohibitionary or exclusionary rules, so long as they are included within its SIP. Finally, the SIP could be amended to reflect special, source-specific limitations. See Memorandum, Options for Limiting the Potential to Emit (PTE) of a Stationary Source Under Section 112 and Title V of the Clean Air Act, at 2-4 (EPA Jan. 25, 1995).

For each of these regulatory methods, however, EPA has proposed conditions for achieving “federal enforceability” that go beyond the mere effectiveness of particular constraint as a practical matter. Inclusion in the SIP, for example, is required in each instance even though EPA’s own approach suggests that it is a consideration independent of and in addition to the need that a constraint be effective for it to count towards reductions. There may, moreover, be regu*1364latory techniques in addition to those that EPA deems susceptible to “federal enforceability” that are equally effective, and yet which are foreclosed as mechanisms for reducing a source’s capacity to emit as a result of EPA’s approach.

What EPA has not explained is how its refusal to consider limitations other than those that are “federally enforceable” serves the statute’s directive to “consider[] controls” when it results in a refusal to credit controls imposed by a state or locality even if they are unquestionably effective. Under EPA’s regime, even a state program of unassailable effectiveness would not qualify in computing a source’s capacity to emit unless it had been submitted not only for EPA approval, but also for inclusion in the SIP. In doing so, EPA would sacrifice a statutory objective in pursuit of ends that, at least as presented in argument to us, have not been justified, either in terms of § 112 or other provisions of the Act. EPA has not explained why it is essential that a control be included within a SIP. It is not apparent why a state’s or locality’s controls, when demonstrably effective, should not be credited in determining whether a source subject to those controls should be classified as a major or area source.20

The government contends that its interpretation of § 112 accords with legislative history. The Senate version of the bill described controls as “installed and operating,” S.16B0, 101st Cong., 1st Sess. § 112(a) (1989); see also S.Rep. No. 228, swpra, at 151, and that version was rejected. But the term “installed and operating” controls presumably included controls imposed by an operator whether or not mandated by any regulations. The Senate had apparently wished to focus exclusively on the actual operating conditions. That Congress rejected that approach does not necessarily suggest that it implicitly delegated to EPA authority to limit the class of govemmentally imposed restraints that could be taken into account to those that are “federally enforceable.”

The government’s argument that Congress, when it enacted the amendments, implicitly ratified EPA’s past treatment of non-federally enforceable controls is not persuasive either. As we have indicated, throughout the 1980s EPA vacillated on the degree of federal involvement required. And when Congress voted in 1990, the 1989 regulations requiring federal enforceability as a prerequisite to recognition were the subject of litigation challenging their legality under the existing statute. Given this situation, Congress cannot be said to have ratified EPA’s position by introducing the phrase “considering controls” without an explicit “federal enforceability” limitation.

As presented in its briefs, EPA’s core justifications for its federal enforceability policy are the need to avoid the administrative burden that EPA would have to bear were it obligated to evaluate the effectiveness of state and local controls and the desirability of uniformity in environmental enforcement. These, of course, are not illegitimate agency objectives. Administrative problems, in particular, can under certain circumstances inform an agency’s construction of imprecise statutory language. See Drummond Coal Co. v. Hodel, 796 F.2d 503, 507 (D.C.Cir.1986). Here, however, EPA would have us accept a rather strained interpretation of the statute based on what appears to be only its unwillingness to evaluate any state or local controls that are not federalized. If there is a closer fit between the notion of “federal enforceability” and § 112’s concern with crediting effective controls it is not evident on this record.

As for national uniformity, the government contends that “one of Congress’ driving concerns in amending the hazardous air pollutants provision in the Act in 1990 was to *1365remedy the haphazard state of air toxic regulations .... The states’ approaches to regulation varied widely,” creating “ ‘a patchwork of differing standards’ ” (citing H.R.Rep. No. 490(1), 101st Cong., 2d Sess. 232 (1990)). Just so; but the amendments do create a national substantive standard, namely categories of sources (major and area) and corresponding technological compliance measures. By no means does that suggest that Congress necessarily intended for state emissions controls to be disregarded in determining whether a source is classified as “major” or “area” under that national standard. Nor did Congress mandate that EPA assume the administration and enforcement of all governmental efforts at emissions limits. If such administration and enforcement is necessary to ensure that controls are effective in the context of the extant regulatory environment, EPA has certainly not made that case and has not indicated how that consideration supports its claim that its interpretation of the statute is reasonable.

In sum, EPA’s definition of “major source” without respect to source categories or two-digit SIC codes is reasonable, as is its requirement that fugitive emissions be included in a source’s aggregate emissions in determining whether the source is major. We therefore deny the petition for review with respect to these issues, advanced here by petitioners General Electric, National Mining Association and American Forest and Paper Association. However, EPA has not explained why the criteria for federal approval and the consequences of that approval are related to ensuring the practical effectiveness of state controls such that the set of controls considered under § 112 should be limited in that fashion. We therefore grant the petition for review with respect to the challenge raised by Chemical Manufacturers Association and American Petroleum Institute.

PILLARD, Circuit Judge:

In this case we consider a pair of challenges to a 2012 regulation promulgated by the U.S. Environmental Protection Agency (EPA), revising Clean Air Act standards for emissions of hexavalent chromium. Hexavalent chromium is a carcinogenic compound emitted into the air during various chrome-finishing processes at more than a thousand facilities across the country. The facilities regulated under the challenged rule provide final, shiny, corrosion-resistant coatings on the *4surfaces of products ranging from plumbing fixtures to airplane wings. The various finishing processes used at those facilities all have the unfortunate side effect of generating misty chromium emissions that, if not properly controlled, can cause cancer. Facilities limit those emissions through the use of devices that capture emissions from the -finishing tanks, or with fume suppressants that inhibit chromium droplets from bursting from the tank sur.face into the air in the first, place. The new rule imposes more stringent emissions limitations than its predecessor and mandates the phase-out'of a category of fume suppressants containing the toxic compound perfluorooctyl sulfonate (PFOS).

Various environmental organizations and an industry association have filed petitions challenging EPA’s revised rule. The environmental petitioners — the Clean Air Council, California Communities Against Toxics, and the Sierra Club — argue that the rule is too lax because EPA ignored relevant information and impermissibly considered costs in calculating revised emissions standards. The industry petitioner — the National Association for Surface Finishing (the Association) — argues that the rule is too stringent. The Association contends that EPA failed to make a determination of developments in practices, processes, or control technologies that the Association claims is a statutorily required precondition to rule revision, that the agency lacked adequate support in the record for phasing out the PFOS-based fume suppressants, and that EPA unreasonably assessed public health risk. The environmental petitioners and the Association intervened in each other’s cases, and we consolidated the cases for review. We deny the petitions.

I.

Section 112 of the Clean Air Act requires EPA to promulgate, and periodically revise as appropriate, national emissions standards for hazardous air, pollutants. See 42 U.S.C. § 7412(d). When Congress enacted that emissions standards program in 1970, it directed EPA to identify and regulate hazardous air pollutants. Dissatisfied with EPA’s' progress in identifying hazardous air, pollutants, Congress amended the Act in 1990 to name nearly 200 such pollutants, including chromium compounds, and charged EPA with identifying sources of those pollutants and setting emissions standards for them. See 42 U.S.C. § 7412(b)(1), (c), (d); see also, e.g., Mexichem Specialty Resins, Inc. v. EPA, 787 F.3d 544, 549-50 (D.C.Cir.2015); Natural Res. Def. Council v. EPA (“NRDC”), 529 F.3d 1077, 1079 (D.C.Cir.2008). EPA undertakes two basic regulatory tasks under section 112 that are relevant to this case: initial promulgation, followed by periodic review and potential revision, of emissions standards.

EPA promulgates an emissions standard for a given pollutant by first determining the average emissions already achieved by the top tier of least polluting emitters, then considering whether a more demanding standard might be practicable and cost effective, and, if so, setting a standard that pushes beyond current practice. For starters, the agency identifies the 12% of facilities that emit the pollutant at the lowest levels, and then calculates the average level of emissions achieved by those facilities. 42 U.S.C. § 7412(d)(3). That calculation is dubbed the “MACT floor” because it is based on “maximum achievable control technology,” and the standard EPA promulgates must not be less stringent than that performance “floor.” See Mexichem, 787 F.3d at 549-50 & n. 2; Nat’l Lime Ass’n v. EPA, 233 F.3d 625, 629 (D.C.Cir.2000). EPA then considers whether, taking into account costs, certain *5health and environmental effects, and .energy requirements, a standard might be practicable that would go beyond the emissions reductions those existing facilities already achieve. See 42 U.S.C. § 7412(d)(2); Nat’l Lime, 233 F.3d at 629. If a more stringent standard is practicable in view of those factors, EPA promulgates a “beyond-the-floor” standard at that more stringent level; otherwise the agency sets the standard at the performance-based MACT floor. See Mexichem, 787 F.3d at 549-50.

EPA then periodically reviews and, if appropriate, revises the promulgated emissions standard, starting within eight years of the initial promulgation. That entails two distinct, parallel analyses: a recurring “technology review” under section 112(d)(6) and a one-time “risk review” under section 112(f)(2). In the technology review, EPA periodically assesses, no less often than every eight years, whether standards should be tightened in view of developments in technologies and practices since the standard’s promulgation or last revision, and, in particular, the cost and feasibility of developments and corresponding emissions savings. See 42 U.S.C. § 7412(d)(6); see also Ass’n of Battery Recyclers, Inc. v. EPA (“ABA”), 716 F.3d 667, 673-74 (D.C.Cir.2013).

Separately, in the one-time risk review, EPA addresses, within eight years of a standard’s promulgation, lingering public health risk that the initial standard did not eliminate. See 42 U.S.C. § 7412(f)(2). To that end, EPA first considers whether the residual health risk is “acceptable,” a threshold EPA generally interprets as carrying cancer incidence no greater than 100 in one million. 75 Fed.Reg. 65,068, 65,07172 (Oct. 21, 2010); see NRDC, 529 F.3d at 1082. If the risk is not acceptable, EPA sets a more stringent standard regardless of cost to bring the risk down to an acceptable level. Even if a risk would be deemed acceptable because it is under that threshold, however, EPA considers whether a more stringent standard is “required in order to provide an ample margin of safety to public health.” 42 U.S.C. § 7412(f)(2)(A). What determines whether a more stringent standard providing an ample margin of safety is statutorily required is the agency’s consideration of health information as well as costs, economic impact, feasibility, and other relevant factors. See id.; see also NRDC, 529 F.3d at 1083; 75 Fed.Reg. at 65,072. If, taking those relevant considerations into account, further risk reductions are attainable beyond the “acceptable” threshold, they are required.

II.

Two decades ago, EPA promulgated standards restricting emissions of hexava-lent chromium from the types of chromium electroplating and anodizing facilities regulated under the rule challenged in this case.1 60 Fed.Reg. 4,948 (Jan. 25, 1995). The agency has reviewed and revised those standards over the years. At issue here is the 2012 Final Rule that most recently updated them. See Final Rule, National Emissions Standards for Hazardous Air Pollutant Emissions: Hard and Decorative Chromium Electroplating and *6Chromium Anodizing Tanks; and Steel Pickling — HC1 Process Facilities and Hydrochloric Acid Regeneration Plants, 77 Fed.Reg. 58,220 (Sept. 19, 2012).

The initial, 1995 rule set emissions limits and allowed facilities to demonstrate compliance with them either by directly measuring chromium emissions or by gauging the surface tension of the chemical baths in their finishing tanks. 60 Fed.Reg. at 4,953-54, 4,956, 4,959. The latter compliance option worked because of the “direct link between surface tension and emissions.” Id. at 4,959. As we explain below, surface tension describes how strongly a fluid sticks together where it meets the air. The surface tension of the chromium solution in which products are submerged during the finishing process determines the force with which non-chromium gas bubbles generated in the process burst out of the solution in the finishing tanks. Reducing surface tension makes it easier for bubbles to pass through the solution into the air, which diminishes the force of bubble-bursting at the surface. That, in turn, reduces the amount of harmful chromium droplets that pop into the air and create misty toxic emissions. In a 2004 rulemaking, EPA reaffirmed and modified the option of compliance by controlling surface tension. 69 Fed.Reg. 42,885, 42,886-88 (July 19, 2004).2

Before issuing the Final Rule petitioners challenge, EPA published a 2010 notice of proposed rulemaking (Notice) and a 2012 supplemental notice of proposed rulemak-ing (Supplemental Notice). In the initial Notice, EPA proposed to determine that the data before the agency did not warrant tightening the existing emissions standards. 75 Fed.Reg. at 65,093-94. EPA nonetheless solicited additional data and comments. Id. at 65,125. Meanwhile, EPA proposed to phase out surfactant-based chemical fume suppressants containing the hazardous chemical PFOS. Id. at 65,094. Surfactants reduce the surface tension of solutions to which they are added. EPA concluded that less toxic, non-PFOS-based fume suppressants had proven effective at controlling surface tensions, even though they were not yet used extensively in the industry. The agency saw no reason why non-PFOS-based suppressants could not cost-effectively replace their toxic PFOS-containing counterparts, so the agency solicited input on its proposal to prohibit PFOS. Id.

Nearly a year and a half later, after having received more accurate and comprehensive data in response to the Notice, EPA in the Supplemental Notice proposed to reverse its prior tentative determinations that more stringent emissions standards were unnecessary. 77 Fed.Reg. 6,628 (Feb. 8, 2012). EPA suggested that revised standards were warranted based on both its technology review and its risk review. Id. at 6,631-52. In its technology review, EPA noted its additional analyses based on new data and explained its proposed selection of feasible, cost-effective options to further limit emissions. Id. at 6,632, 6,638-45. EPA solicited information regarding whether the surface tension limits proposed in the Supplemental Notice were achievable using non-PFOS-based suppressants. Id. at 6,645. In its risk review, EPA proposed to conclude that cancer risk was well below the “acceptable” 100-in-one-million level, and that the same standards the agency was proposing as a result of the technology review would likewise provide the requisite “ample margin of safety” for purposes of its risk review. Id. at 6,648-49.

*7EPA issued the Final Rule in September 2012, adopting the emissions standards proposed in the Supplemental Notice. 77 Fed.Reg. at 58,225. EPA explained that additional information submitted during the interim period and additional analyses the agency had performed bolstered the determinations proposed in the Supplemental Notice. Id. at 58,225-26. EPA also finalized the phase-out of PFOS-based fume suppressants, emphasizing data the agency collected from Minnesota facilities that had been controlling surface tensions effectively using non-PFOS-based suppressants. Id. at 58,230, 58,236-37.

III.

This court applies the familiar, deferential standard announced in Chevron, U.S.A., Inc. v. Natural Resources Defense Council, Inc., to sustain any reasonable agency interpretation of ambiguity in the Clean Air Act. 467 U.S. 837, 842-44, 104 S.Ct. 2778, 81 L.Ed.2d 694 (1984). We reverse EPA’s determinations under the Act when they are “arbitrary, capricious, an abuse of discretion, or otherwise not in accordance with law.” 42 U.S.C. § 7607(d)(9)(A). Under arbitrary-and-capricious review, EPA’s determinations are “presumptively valid provided [they] meet[] a minimum rationality standard.” Natural Res. Def. Council, Inc. v. EPA, 194 F.3d 130, 136 (D.C.Cir.1999). We uphold EPA’s determinations so long as “EPA acted within its delegated statutory authority, considered all of the relevant factors, and demonstrated a reasonable connection between the facts on the record and its decision.” Ethyl Corp. v. EPA, 51 F.3d 1053, 1064 (D.C.Cir.1995). We afford special deference “where the agency’s decision rests on an evaluation of complex scientific data within the agency’s technical expertise.” Troy Corp. v. Browner, 120 F.3d 277, 283 (D.C.Cir.1997).

The environmental petitioners challenge the Final Rule as too lax on the grounds that EPA unreasonably (1) declined to recalculate a technology-based emissions stringency floor that is independent of cost considerations, and (2) ignored emissions data from California in calculating revised standards. The Association challenges the rules as unauthorized and overly stringent on the grounds that EPA (1) did not adequately tie its revision of standards to technological “developments” that had occurred since the initial standard-promulgation, (2) concluded without sufficient evi-dentiary support that non-PFOS-based fume suppressants are capable of achieving the rule’s emissions limits, and (3) in its risk review, unreasonably failed to determine whether existing emissions limits provided the requisite margin of safety, deviated from the agency’s prior determination that the requisite margin of safety had been attained, and relied on inaccurate data.

A.

The environmental petitioners contend that EPA was required to calculate a new' MACT floor when it revised emissions standards pursuant to its technology review under section 112(d)(6). Subpara-graph (d)(6) requires EPA, no less often than every eight years, to “review, and revise as necessary (taking into account developments in practices, processes, and control technologies) emission standards promulgated under” that section. 42 U.S.C. § 7412(d)(6). The environmental petitioners assert that EPA’s revision of standards in its technology review must be treated as a “promulgation” of an emissions standard, such that it triggers the Clean Air Act’s requirement that any promulgation must begin with EPA setting a MACT floor. They argue, in effect, that EPA must calculate a new MACT floor whenever the agency revises an existing *8emissions standard based on its periodic technology review under section 112(d)(6). A new MACT floor, if EPA set one, would be based purely on the achievements of the best-performing facilities in the industry after initial emissions standards had been operative for several years, and thus presumably would be more stringent than the existing MACT floor, and likely also more stringent than technology or risk-based revisions that take costs and other potentially constraining factors into account.

EPA itself has not so read the statute, and our precedent binds us to reject the environmental petitioners’ argument that it must so read it.3 We initially confronted a version of this question in NRDC, a case in which EPA had conducted ' a technology review but determined that revision was not warranted. 529 F.3d at 1080. The parties in NRDC briefed the argument the environmental petitioners press here. We concluded that the agency was not required to recalculate the MACT floor at the outset of its technology review, at least where EPA had decided not to revise emissions standards as a result of that review. Id. at 1084.

We had occasion to address the issue more squarely in ABR, a case in which EPA had not only reviewed, but also revised, standards after a technology review. 716 F.3d at 670. The court in ABR was not persuaded by petitioners’ contention— pressed again by the environmental petitioners here — that our decision in NRDC was inapposite on the ground that it dealt with an EPA review that yielded no revision of any standard. ABR emphasized that NRDC rested “on two independent conclusions,” and that in such a case, “the ruling on neither is obiter [dictum], but each is the judgment of the court, and of equal validity with the other.” Id. at 673 (internal quotation marks omitted). We held in ABR that the petitioners’ argument that the agency must recalculate the MACT floor, “although far better developed than the identical claim in [NRDC], [was] barred by that decision.” Id.4

The environmental petitioners incorrectly suggest that our decisions on this point in ABR and NRDC were abrogated by our later decision in White Stallion Energy Ctr., LLC v. EPA, 748 F.3d 1222 (D.C.Cir.2014), which the Supreme Court recently partially reversed in Michigan v. EPA, 576 U.S. —, 135 S.Ct. 2699, 192 L.Ed.2d 674 (2015).5 In White Stallion, we addressed EPA’s interpretation of Clean Air *9Act section 112(n)(l)(A), and its relationship to section 112(d). 748 F.3d at 1242-44. Section 112(n)(l)(A) provides for EPA regulation of fossil-fuel-fired power plants when “appropriate and necessary.” 42 U.S.C. § 7412(n)(1)(A). We sustained as reasonable EPA’s interpretation of “under this section,” as used in that provision, to refer to the entirety of section 112, such that the regulation required whenever section 112(n)(1)(A)’s “appropriate and necessary” criteria are met must comport with the procedures of the rest of section 112, including those of section 112(d), addressing emissions standards generally. 748 F.3d at 1243-44; see 42 U.S.C. § 7412(d). The environmental petitioners assert that the relationship between section 112(n)(l)(A), with its “regulate” as “appropriate and necessary” trigger, and section 112, governing hazardous air pollution regulation generally (including the setting of MACT floors under section 112(d)(2)-(3)), is the same as the relationship between section 112(d)(6), with its “review, and revise as necessary” language, and section 112(d)(2)-(3). However, White Stallion did not tie section 112(n)(l)(A) to section 112(d)(2)-(3) specifically, as the petitioners’ analogy assumes. Nor did that decision address the distinction EPA makes here between initial promulgation under section 112(d)(2)-(3) and promulgation of revised standards as necessary under section 112(d)(6). We simply deferred to EPA’s view that regulation triggered under section 112(n)(l)(A) must follow the procedures of section 112(d).

B.

The environmental petitioners also contend that EPA unreasonably disregarded,' in both its technology and risk reviews, the emissions reductions and technological advancements achieved in California, where that state’s more stringent emissions standards require facilities to employ more ambitious controls.

EPA took account of the California emissions data. In its technology and risk reviews, EPA estimates actual nationwide emissions and costs by extrapolating data collected from a sample of facilities across the country in order to establish baselines against which the agency can assess emissions decreases, health risk, and cost effectiveness associated with different practices, technologies, and emissions limitations. See, e.g., 77 Fed.Reg. at 6,631-34. EPA did not ignore California’s emissions data; it omitted it only from the dataset from which the agency extrapolated other states’ emissions. Id. at 6,633-34. That was prudent as a matter of statistical accuracy because “California plants are not representative of emissions for non-California plants.” Id. at 6,634; see also J.A. 531 (same in response to comments). EPA used California data, meanwhile, to estimate emissions for other plants in that state. 77 Fed.Reg. at 6,634. The agency thus “did not exclude the California [emissions] data from the overall analysis,” but rather “treated the data from plants in California differently” so as to achieve a statistically accurate portrait of nationwide emissions. Id. EPA’s approach was .reasonable because extrapolating California’s nationally unrepresentative data to the rest of the country would have distorted the emis*10sions figures that EPA uses in its cost-effectiveness and risk analyses, whereas reliance on that data for California-specific estimates had no such distorting effect.

EPA also did not ignore California’s technological advances in assessing the cost effectiveness of applying nationally the standards in effect in California. The agency acknowledged that California emissions standards require controls that are more stringent than those required elsewhere, and that they had been feasibly implemented in California. The agency weighed the costs and benefits of requiring those controls nationally, and concluded that the relatively minimal reductions in emissions and risk that could be expected from extending California-like controls across the country would not warrant the associated expense. EPA “considered the option of requiring controls similar to standards adopted in California, which would essentially require facilities to install high efficiency particulate air (HEPA) filters,” but found that “overall costs for that option were significantly higher than ... other options” EPA had examined, “and would have resulted in much greater economic impacts to small businesses.” 77 Fed.Reg. at 58,227.

The agency further explained that “requirements similar to the California standards” were “not appropriate ... to provide an ample margin of safety to protect public health” in light of “the high overall costs and economic impacts.” Id. Responding to a comment that EPA had “failed to provide any explanation for not considering the California reductions,” the agency explained that it had “evaluated ... the controls used to comply with the standards in California” and concluded “that requiring these controls throughout the industry was not appropriate under either section 112(d)(6) [technology review] or section 112(f)(2) [risk review].” Id. at 58,231-32; see also 75 Fed.Reg. at 65,092-94 (assessing risk level and cost associated with HEPA filters). The environmental petitioners suggest that EPA should have performed additional analyses beyond that related to HEPA filters but, as the agency reasonably explained, EPA lacked the necessary information to do so: EPA was not provided with, and was otherwise unaware of, the detailed explanations of the assumptions and methodology behind California cost analyses that the agency would have needed to parse California’s data and accurately assess its applicability to the national regulatory context. See, e.g., J.A. 543.

EPA took into account the statutorily required considerations of, inter alia, cost, emissions reductions, and health risk. The agency then provided a transparent, reasoned explanation of its decisions, considering all relevant information in the record. The statute does not mandate a particular method of cost-benefit analysis. Therefore, we defer to EPA’s methodology as well as its ultimate balancing decisions. See Husqvarna AB v. EPA, 254 F.3d 195, 200 (D.C.Cir.2001); cf. Brown v. United States, 327 F.3d 1198, 1205-06 (D.C.Cir.2003) (deferring to an agency’s reasoned methodology even under the less deferential Skidmore standard of review).6

C.

The Association, in the first of its three challenges, asserts that EPA unreasonably *11determined in its technology review that “developments” had occurred after the original rulemaking that required revision of the existing emissions standards. The statute calls on EPA to revise promulgated standards “as necessary (taking into account developments in practices, processes, and control technologies).” 42 U.S.C. § 7412(d)(6). EPA and the Association agree that the agency could not revise standards in its technology review unless such pertinent “developments” occurred after the 1995 rule. The Association argues, however, that EPA failed to identify the requisite “developments” and failed to specify how its revised standards were tied to them. The Association also suggests EPA was arbitrary and capricious in changing course: in 2010 its proposed determination was that “developments” did not warrant revision, but in 2012 EPA concluded that revision was warranted after all. The Association further argues that EPA’s statistical analysis of emissions data was unreasonably flawed.

EPA permissibly identified and took into account cognizable developments in practices, processes and control technologies. The agency interprets “developments” to encompass not only wholly new methods, but also technological “improvements ... that could result in significant additional emission reduction.” 75 Fed.Reg. at 65,-083. “Developments” thus include “improvements in efficiency, reduced costs or other changes that indicate that a previously considered option for reducing emissions may now be cost effective or technologically feasible.” 77 Fed.Reg. at 58,231.

The Association does not directly challenge those interpretations, but maintains that EPA failed to identify what specific “developments,” including technological “improvements,” occurred in this case that warranted revision. EPA identified several technologies, such as emissions elimination devices, HEPA filters, enclosing tank hoods, and fume suppressants. 77 Fed. Reg. at 6,632 (2012 Supplemental Notice referencing developments identified in 2010 Notice); 75 Fed.Reg. at 65,083, 65,093-94 (Notice identifying the aforementioned developments); J.A. 152-54 (2010 technology review memorandum discussing the same). EPA further explained that, although the technologies were not brand new, improvements in the performance of some of them had resulted in emissions reductions. J.A. 516 (2012 response to comments); see also 77 Fed.Reg. at 6,631-32 (acknowledging data showing improved facility performance).

It suffices for EPA to assess and discuss the collective impact of the developments it has identified, and to revise standards appropriately in light thereof. Section 112(d)(6) simply requires that EPA “take[ ] into account developments in practices, processes, and control technologies” when considering whether to revise standards. The agency explained that it had examined what emissions levels could be achieved using various add-on control devises and fume suppressants, including developments the agency had previously identified. See, e.g., 77 Fed.Reg. at 6,632. EPA went on to provide details regarding the costs and emissions reductions identified, and evaluated in the course of arriving at its conclusion that specified, best-available control technologies could cost-effectively meet more stringent emissions standards. See, e.g., id. at 6,640. The statute does not require EPA to identify a nexus between each distinct development and the revised standards. EPA’s decision-making was sufficiently clear and rational. See Int’l Fabricare Inst. v. EPA, 972 F.2d 384, 389 (D.C.Cir.1992).

Moreover, the shift in EPA’s position from 2010 to 2012 was reasonable because the agency received intervening informa*12tion relevant to its decision. See, e.g., Sierra Club v. EPA, 353 F.3d 976, 986-87 (D.C.Cir.2004) (“[T]he adequacy of the underlying justification offered by the agency is the pertinent factor — not what the agency did on a different record.... ”). In 2010, EPA proposed to leave the existing emissions standards in place, not because no technological developments had occurred, but because the agency determined based on the data before it that requiring nationwide implementation of those technologies was not feasible, cost effective, and otherwise warranted. 75 Fed.Reg. at 65,093-95. EPA’s Notices solicited comments, however, and EPA contacted pollution control authorities to gain fuller, more accurate data. New data showed that facilities were performing better with respect to emissions than previously thought. 77 Fed.Reg. at 58,225-26, 58,231; 77 Fed.Reg. at 6,631-32. Reductions in emissions are, of course, relevant to the cost effectiveness of emissions-control technologies in controlling emissions. See, e.g., J.A. 483-89 (2012 memorandum). The agency thus concluded, based on the improved data, that more stringent technology-based standards were cost effective and otherwise appropriate. See 77 Fed.Reg. at 58,225-26; 77 Fed.Reg. at 6,632-33, 6,63845. EPA’s change in position therefore was not arbitrary and capricious.

Finally, the dataset EPA used to estimate facilities’ compliance rates — a component of the agency’s technology review— was not impermissibly small or inaccurate. The Association charges that EPA’s analysis was not designed to provide an accurate cross-section of the industry. It asserts that EPA did not sample facilities “on any statistical or representative basis,” and faults the agency for not requesting “all emissions data the states had in their files.” Ass’n Br. 30. EPA responds that it relied on data from a statistically significant number of facilities and a range that reflected “current, available information from facilities across all major regions of the country.” EPA Br. 50. Reacting to concerns raised after the 2010 Notice that the agency’s dataset was too small and unrepresentative, EPA sought detailed information about emissions associated with various facility types. 77 Fed.Reg. at 6,631. EPA explained that emissions data available for collection is limited because many smaller facilities are not required to submit annual emissions inventories — an informational dearth in light of which EPA gathered what it concluded was- the “best information available to the Agency” and “minimized the [statistical] uncertainties to the extent feasible.” J.A. 535-36. In the face of those challenges, the agency obtained information from 24 state and local agencies, as well as some information from the Association itself, to achieve a sample that included data from 301 operational plants. 77 Fed.Reg. at 6,631. EPA determined that conducting a more in-depth industry-wide survey, requiring more facilities to collect and submit additional data, would have been statistically unnecessary and would not have warranted the substantial burden on smaller facilities. J.A. 537.

Tellingly, the Association does not identify any specific, superior statistical threshold of data-gathering comprehensiveness that, in its view, would have been reasonable. Rather, acknowledging that a limited amount of emissions data is available to EPA, the Association simply criticizes the agency for not obtaining and evaluating more data. But “[w]e generally defer to an agency’s decision to proceed on the basis of imperfect scientific information, rather than to invest the resources to conduct the perfect study.” NRDC, 529 F.3d at 1086 (quoting Sierra Club v. EPA, 167 F.3d 658, 662 (D.C.Cir.1999)). We recognize that EPA conceiv*13ably “could have used better data in conducting its risk analysis,” but that “misstates [our] inquiry under the arbitrary and capricious standard,” which is “whether EPA has acted reasonably, not whether it has acted flawlessly.” Id. (emphasis omitted). We cannot say that the size and representativeness of the dataset EPA used to estimate compliance rates was statistically unreasonable, nor can we characterize as arbitrary EPA’s considered judgment that collecting additional data was unnecessary and not worth the cost, keeping in mind the “wide latitude” we afford the agency in its data-gathering decisions. Id. (quoting Sierra Club, 167 F.3d at 662); see also White Stallion, 748 F.3d at 1247-48 (“EPA’s data-collection process was reasonable, even if it may not have resulted in a perfect dataset.”), rev’d in part on other grounds, Michigan, supra. We are satisfied that EPA’s data-gathering and analysis was adequate.

D.

The Association further argues that EPA’s decision to phase out PFOS-based fume suppressants was arbitrary and capricious. As we explain below, fume suppressants reduce finishing baths’ surface tension and, consequently, their chromium emissions. The Association’s claim hinges on its contention that there is inadequate support in the record showing that non-PFOS-based substitutes can effectively achieve the emissions limits required by the Final Rule. The Association does not challenge the compliance option that facilities have enjoyed since the original 1995 rule to measure the surface tension of their finishing baths rather than the emissions therefrom. See 60 Fed.Reg. at 4,953-54, 4,956. Nor does it seriously question that the administrative record shows that non-PFOS-based suppressants are capable of meeting the Final Rule’s surface tension limits. Rather, the Association contends that EPA failed empirically to support the conclusion that the same emissions control will result from equal surface tension reductions, whether those reductions are achieved with non-PFOS-based suppressants or PFOS-based suppressants. To that end, the Association points to what it perceives as an absence of reliable studies that measure the emissions from baths whose surface tension is controlled by non-PFOS-based suppressants.

The Association’s challenge fails to appreciate the finding, documented in the record, that the relationship between surface tension and emissipns does not depend on the identity of the suppressant used to achieve that surface tension. Reducing a finishing bath’s surface tension reduces emissions from that bath. Surface tension describes a liquid’s cohesive strength where the liquid meets the air. The higher the surface tension, the more the liquid will bulge when its surface is pressured and the more resistant it will be to allowing bubbles to break through. E.g., J.A. 349. That is significant in the context of surface finishing because the electrolysis-induced finishing process generates bubbles of gas within the finishing solution. When those bubbles fizz up through the chromium-solution bath and pop upon hitting the surface, the bursting action sends tiny droplets of the solution into the air as mist. J.A. 440. Adding a surfactant-based fume suppressant to the finishing solution reduces its surface tension, which allows gases generated within the bath to pass through the surface of the solution more easily. That results in less forceful bubble-popping at the surface, which in turn diminishes the amount of chromium mist emitted. E.g., 77 Fed.Reg. at 58,236-37; J.A. 349, 440-43. Because the physical relationship between surface tension and emissions exists independent of the specific tension-controlling agent, it *14is sufficient for EPA to rely on the evidence in the record establishing the relationship between non-PFOS-based suppressants and surface tension; EPA need not identify studies directly confirming the relationship between the specific surface tension controlling agent and the emission reduction.

That physical relationship between surface tension and emissions level is the basis of EPA’s long-standing rule allowing facilities to demonstrate compliance with emissions standards by measuring finishing baths’ surface tension rather than directly measuring their emissions. It is the established “direct link between surface tension and emissions” that has supported EPA’s acceptance of surface tension control as a proxy for emissions control. 60 Fed.Reg. at 4,959 (1995 rule).

The 1995 rulemaking in which EPA determined to rely on the relationship between surface tension and emissions never suggested that the type of fume suppressant used to control surface tension matters to emissions level. See id. (“Based on data collected by the EPA, the [emissions] performance of an electroplating bath controlled with a wetting agent-type [i.e. chemical surfactant] fume suppressant can be determined by the surface tension of the bath.”); see also 58 Fed.Reg. 65,768, 65,795 (Dec. 16, 1993) (original proposed rule) (“Measurement of the surface tension of the bath ... at least once every 4 hours during operation of the tank would be sufficient to ensure continuous compliance with the emission limit.”).

EPA explicitly confirmed in the 2004 rulemaking that the tension/emissions relationship does not depend on suppressant type when it recognized that “[s]ources will be in compliance with the emission limits provided the surface tension is maintained at or below the proposed limits, regardless of the type of fume suppressant used”; the data before the agency did not “indicate that emission control levels are a function of the type of fume suppressant used in the tank solution.” 69 Fed.Reg. at 42,888.

EPA reiterated in 2012 that “emissions are a function of the surface tension” and that lower surface tension results in lower emissions “regardless of the specific chemicals used.” J.A. 479 (EPA memorandum cited in Final Rule); see also 77 Fed.Reg. at 58,237 (“It is our understanding that this relationship between surface tension and chromium emissions is dependent primarily on the surface tension of the tank and not on the product used to reduce surface tension.”). It is the property of surface tension on the bath, not some oth,er reaction unique to the particular surfactant employed, that affects the force with which bubbles in the chromium bath burst and send droplets into the air. See, e.g., J.A. 349, 44043.

Neither logic nor the record supports the Association’s conclusory suggestion that emissions will vary, given a constant surface tension, depending on whether PFOS- or non-PFOS-based fume suppressants are used to achieve that surface tension. The Association points out that EPA’s statements in earlier rulemakings were made in the context of considering only various PFOS-based suppressants. That ignores the fact that EPA restated in 2012, in a memorandum in which it was also considering the effectiveness of non-PFOS-based suppressants, that the suppressant’s chemical identity is irrelevant. See J.A. 479, 482. Moreover, the Association provides no explanation and points to no evidence supporting its suggestion that the principle EPA earlier recognized does not apply to non-PFOS-based suppressants. If EPA had established, in a rule-making involving various kitchen stoves, that water boils when it reaches 212°F *15regardless of the type of stove used, and later took that boiling point as a given in a rulemaking involving campfires, we would not require EPA to document the same scientific phenomenon of water boiling at 212°F in the new, campfire context — at least not where a petitioner proffered nothing suggesting why the heat-source difference might matter. The Association’s bare assertion that EPA needed to provide fresh empirical support for the applicability of the tension/emissions relationship in the context of non-PFOS-based suppressants is insufficient to undermine the documentation already in the record. “Agencies do not need to conduct experiments in order to rely on the prediction that an unsupported stone will fall.” Assoc’d Gas Distribs. v. FERC, 824 F.2d 981, 1008-09 (D.C.Cir.1987); cf. Chamber of Commerce of U.S. v. SEC, 412 F.3d 133, 142 (D.C.Cir.2005) (“[W]e are acutely aware that an agency need not — indeed cannot — base its every action upon empirical data....”).

The studies in the administrative record are consistent with EPA’s understanding that surface tension is a reliable proxy for emissions regardless of the fume suppressant employed.- A 2011 study expressly contemplates the PFOS/non-PFOS dichotomy and treats that difference as immaterial to the tension-emissions relationship. See J.A. 439-49 (Neil Patton & Gene Barlowe, Atotech, Non-PFOS, Permanent Mist Suppressants for Hard Chromium Plating, Decorative Chromium Plating and Chromic Etch Applications (2011)). The study identifies “mist suppressants” as a general category of surfactants, noting they all “work by reducing the surface tension of the solution.” J.A. 442; see id. (stating without reference to the surfactant’s chemical composition that reducing surface tension lowers mist generation). The study then identifies several characteristics of effective suppressants — including. the ability to control surface tension and emissions, as well as unrelated criteria such as effect on plating quality — without suggesting that ability to meet those criteria depends on PFOS composition. J.A. 444. The study explains that earlier generations of non-PFOS-based suppressants were disfavored not because of non-correlation between surface tension and emissions, but rather because they were less stable and tended chemically to foment a scum that degraded plating quality. J.A. 445. The authors go on to recognize that newer non-PFOS-based suppressants for years have been performing “just as well” as PFOS-based suppressants. J.A. 444-47. Other studies in the record, which measured only emissions directly and did not document emissions’ relationship to surface tension, are consistent with that understanding.7

Therefore, it suffices for EPA to show that non-PFOS-based suppressants are as effective at controlling surface tension as PFOS-based suppressants (which the Association does not contend are incapable of achieving the Final Rule’s surface tension limits). We conclude that the data from Minnesota facilities emphasized in the Final Rule shows that effectiveness. See 77 *16FecLReg. at 58,236-37. The Minnesota data show that non-PFOS-based suppressants perform, in the facilities’ experience, “pretty similarly” and pose “no noticeable differences” in comparison to PFOS-based suppressants. J.A. 415-16; see J.A. 417-21 (raw data supporting those observations); J.A. 482 (additional data analysis). The Association does not meaningfully argue otherwise, conceding that the Minnesota data “indicate that non-PFOS fume suppressants might be used to meet the new surface tension standards.” Ass’n Br. 41.

E.

The Association argues that EPA’s risk review under section 112(f)(2) was unreasonable, for three reasons.8

First, the Association asserts.that EPA failed to make a finding in the Final Rule that a revised standard was “required in order to provide an ample margin of safety to public health.” 42 U.S.C. § 7412(f)(2). However, EPA did not skip the step of determining in the Final Rule that revised standards were required to provide an ample margin of safety. The agency expressly stated that it was “promulgating standards under Section 112(f) to provide an ample margin of safety.” 77 Fed.Reg. at 58,226; see also id. at 58,229 (“[A]fter considering all the health and cost factors ..., the agency has determined that ... [the revised standard] will provide an ample margin of safety....”).

The Association argues that EPA’s determinations that revised standards would provide an ample margin of safety did not contain a determination that they were required to provide that margin. That is simply not the case. In response to comments, for instance, the agency reiterated its understanding that section 112(f)(2) mandates revision when, “required to provide an ample margin of safety.” J.A. 527. The agency has consistently viewed itself as “required to propose emissions standards ... that reduce risk to a level that is acceptable and provides an ample margin of safety.” 75 Fed.Reg. at 65,091. It is precisely as a result of the further consideration and context-based determination that additional risk reductions can be achieved cost effectively and in light of other statutorily relevant factors that the agency’s obligation arises to tighten standards to provide an “ample margin of safety.” EPA’s determination that revision was required was therefore at least implicit.

Second, the Association argues that any finding EPA may have made in 2012 that the revised standards were required to protect the public health was an arbitrary and capricious reversal of its earlier 2010 determination. The Association contends, more specifically, that it was unreasonable for EPA to conclude in 2010 that the requisite “ample margin of safety” was in place, but to conclude in 2012 that revised standards were necessary to achieve such a margin of safety — and to do so at the same time' that EPA acknowledged that health risks were equal to or lower than it had thought they were in 2010. EPA’s “ample margin of safety” determination does not hinge solely on public health risk, however; it may also take into account “costs and economic impacts, technological feasibility, and other [non-health] factors.” NRDC, 529 F.3d at 1083 *17(internal quotation marks omitted). That being the case, the fact that EPA considered a prior standard to provide an ample margin of safety under older data does not prevent EPA from determining that new data on cost and feasibility require a more stringent standard. In this case, EPA gained significant new data between 2010 and 2012 that changed its calculus about cost effectiveness and feasibility. 77 Fed.Reg. at 58,225-26; J.A. 521-22 (response to comments). EPA’s change in position was therefore reasonable. See Sierra Club, 353 F.3d at 986-87; cf. Nat’l Ass’n of Mfrs. v. EPA, 750 F.3d 921, 925 (D.C.Cir.2014) (“[W]e do not assign ‘presumptive validity’ to the prior [standards]; the question is whether EPA reasonably explains the current standards.”).

Lastly, the Association characterizes EPA’s statistical methodology for estimating emissions (and, consequently, cancer risk) as unreasonably flawed. It asserts that the sample of facilities on which EPA based its emissions estimates in the Supplemental Notice included many facilities that were either closed or not subject to regulation, or that emissions estimates from those -facilities were inaccurately high.

The Association identified inaccuracies in the sample EPA had used by contacting facilities in that sample and comparing the information they provided against EPA’s assumptions. It also reported that those inaccuracies were corroborated by a supplemental sample it generated to check the accuracy of EPA’s model. J.A. 284-305, 312-27 (March 2012 submission in response to Supplemental Notice). The Association claims that its supplemental data show that EPA’s initial sample was unacceptably unrepresentative of the larger population of facilities.

EPA responded by adjusting its data inputs, removing all the data that the Association had identified as coming from closed facilities or those that did not emit hexavalent chromium, and added the more specific, verifiable data that the Association had provided. 77 Fed.Reg. at 58,235; J.A. 546-47 (response to comment); see also J.A. 637-38 (EPA correspondence to the Association further explaining the adjustments). EPA declined, however, to incorporate any of the Association’s data that the agency determined was too vague or lacking in detail to be reliable and susceptible of incorporation into its predictive model. J.A. 638. EPA also declined to adopt the Association’s extrapolations from its supplemental data sample because, the agency explained, the Association did not provide sufficient information about its sampling methodology for EPA to discern whether the supplemental data sample was collected in a statistically representative, reliable fashion. J.A. 647.

The Association has never challenged EPA’s predictive model, as distinct from the data inputs into that model. EPA thus retained the “Monte Carlo” statistical technique of extrapolating an estimate of emissions from all facilities from actual emissions data from a smaller sample of facilities. EPA re-ran its model based on a dataset adjusted as just described, and determined that the resulting emissions estimates supported its risk conclusions. 77 Fed.Reg. at 58,225-30, 58,235-36.

We conclude that EPA’s methodology passes muster. Keeping in mind the “wide latitude” we afford EPA’s expertise-informed choice of data-gathering methodology, we find that EPA’s data-collection process was reasonable. See NRDC, 529 F.3d at 1086 (quoting Sierra Club, 167 F.3d at 662); see also White Stallion, 748 F.3d at 1247-48. The Association contends that EPA’s response to its submission — adjusting the inputs of its dataset— was insufficient. But we must bear in *18mind that “statistical analysis is perhaps the prime example of those areas of technical wilderness into which judicial expeditions are best limited to ascertaining the lay of the land.” Appalachian Power Co. v. EPA, 135 F.3d 791, 802 (D.C.Cir.1998). We uphold EPA’s model as long as the agency “explain[s] the assumptions and methodology used in preparing the model” and “providefs] a complete analytic defense” should the model be challenged. Small Refiner Lead Phase-Down Task Force v. EPA, 705 F.2d 506, 535 (D.C.Cir.1983) (internal quotation marks omitted). As outlined above, EPA provided such an explanation and defense of how and why it formulated the sample and conducted the analysis that it did. See, e.g., 77 Fed.Reg. at 58,235; see also J.A. 467-472 (memorandum specifically explaining and defending EPA’s revised emissions simulation). Moreover, the Association does not explain why the level of uncertainty resulting from EPA’s ultimate sample and analysis was statistically unacceptable, nor does it offer any superior method that it contends EPA should have used instead.

For the reasons above, we deny the petitions for review and uphold the Final Rule.

So ordered.

LEVENTHAL, Circuit Judge:

These consolidated petitions of International Harvester and the three major auto' companies, Ford, General Motors and Chrysler, seek review1 of a decision by the Administrator of the Environmental Protection Agency denying petitioners’ applications, filed pursuant to Section 202 of the Clean Air Act,2 for one-year suspensions of the 1975 emission standards prescribed under the statute for light duty vehicles in the absence of suspension.

I. STATEMENT OF THE CASE

The tension of forces presented by the controversy over automobile emission standards may be focused by two central observations:

(1) The automobile is an essential pillar of the American economy. Some 28 per cent of the nonfarm workforce draws its livelihood from the automobile industry and its products.3

(2) The automobile has had a devastating impact on the American environment. As of 1970, authoritative voices stated that “[ajutomotive pollution constitutes in excess of 60% of our national air pollution problem” and more than 80 per cent of the air pollutants in concentrated urban areas.4

A. Statutory Framework

Congressional concern over the problem of automotive emissions dates back to the 1950’s,5 but it was not until the passage of the Clean Air Act in 1965 that Congress established the principle of Federal standards for automobile emissions. Under the 1965 act and its successor, the Air Quality Act of 1967, the Department of Health, Education and Welfare was authorized to promulgate emission limitations commensurate with existing technological feasibility.6

The development of emission control technology proceeded haltingly. The Secretary of HEW testified in 1967 that *623“the state of the art has tended to meander along until some sort of regulation took it by the hand and gave it a good pull. . . . There has been a long period of waiting for it, and it hasn’t worked very well.” 7

The legislative background must also take into account the fact that in 1969 the Department of Justice brought suit against the four largest automobile manufacturers on grounds that they had conspired to delay the development of emission control devices.8

On December 31, 1970, Congress grasped the nettle and amended the Clean Air Act to set a statutory standard for required reductions in levels of hydrocarbons (HC) and carbon monoxide. (CO) which must be achieved for 1975 models of light duty vehicles. Section 202(b) of the Act added by the Clean Air Amendments of 1970, provides that, beginning with the 1975 model year, exhaust emission of hydrocarbons and carbon monoxide from “light duty vehicles” must be reduced at least 90 per cent from the permissible emission levels in the 1970 model year.9 In accordance with the Congressional directives, the Administrator on June 23, 1971, promulgated regulations limiting HC and CO emissions from 1975 model light duty vehicles to .41 and 3.4 grams per vehicle mile respectively. 36 Fed.Reg. 12,657 (1971).10 At the same time, as required by section 202(b) (2) of the Act, he prescribed the test procedures by which compliance with these standards is measured.11

Congress was aware that these 1975 standards were “drastic medicine,” 12 designed to “force the state of the art.”13 There was, naturally, concern whether the manufacturers would be able to achieve this goal. Therefore, Congress provided, in Senator Baker’s phrase, a “realistic escape hatch”: the manufacturers could petition the Administrator of the EPA for a one-year suspension of the 1975 requirements, and Congress took the precaution of directing the National Academy of Sciences to undertake an ongoing study of the feasibility of compliance with the emission standards. The “escape hatch” provision addressed itself to the possibility that the NAS study or other evidence might indicate that the standards would be unachievable despite all good faith efforts at compliance. This provision was limited to a one-year suspension, which would defer compliance with the 90% reduction requirement until 1976. Under section 202(b)(5)(D) of the Act, 42 U. S.C. § 1857f-l(b) (5) (D), the Administrator is authorized to grant a one-year suspension

only if he determines that (i) such suspension is essential to the public interest or the public health and wel*624fare of the United States, (ii) all good faith efforts have been made to meet the standards established by this subsection, (iii) the applicant has established that effective control technology, processes, operating methods, or other alternatives are not available or have not been available for a sufficient period of time to achieve compliance prior to the effective date of such standards, and (iv) the study and investigation of the National Academy of Sciences conducted pursuant to subsection (c) of this section and other information available to him has not indicated that technology, processes, or other alternatives are available to meet such standards.

The statute provides that an application for suspension may be filed any time after January 1, 1972, and that the Administrator must issue a decision thereon within 60 days. On March 13, 1972, Volvo, Inc., filed an application for suspension and thereby triggered the running of the 60 day period for a decision. 37 Fed.Reg. 5766 (March 21, 1972.)14 Additional suspension requests were filed by International Harvester on March 31, 1972, and by Ford Motor Company, Chrysler Corporation, and General Motors Corporation on April 5, 1972. Public hearings were held from April 10-27, 1972. Representatives of most of the major vehicle manufacturers (in addition to the applicants), a number of suppliers of emission control devices and materials, and spokesmen from various public bodies and groups, testified at the hearings and submitted written data for the public record. The decision to deny suspension to all applicants was issued on May 12, 1972.

The Decision began with the statement of the grounds for denial: “ . . . I am unable, on the basis of the information submitted by the applicants or otherwise available to me, to make the determinations required, by section 202 (b) (5) (D) (i), (iii), or (iv) of the Act.” 15 The EPA Decision specifically focused on requirement (iii) that:

the applicant has established that effective control technology, processes, operating methods, or other alternatives are not available or have not been available for a sufficient period of time to achieve compliance prior to the effective date of such standards • • •>

A Technical Appendix, containing the analysis and methodology used by the Administrator in arriving at his decision, was subsequently issued on July 27, 1972.

B. Initial Decision of the Administrator

The data available from the concerned parties related to 384 test vehicles run by the five applicants and the eight other vehicle manufacturers subpoenaed by the Administrator. In addition, 116 test vehicles were run by catalyst and reactor manufacturers subpoenaed by the Administrator. These 500 vehicles were used to test five principal types of control systems: noble metal monolithic catalysts, base metal pellet catalysts, noble metal pellet catalysts, reactor systems, and various reactor/catalyst combinations.

At the outset of his Decision, the Administrator determined that the most effective system so far developed was the noble metal oxidizing catalyst.16 Addi*625tionally, he stated that the “most effective systems typically include: improved carburetion; a fast-release choke; a device for promoting fuel vaporization during warm-up; more consistent and durable ignition systems; exhaust gas recirculation; and a system for injecting air into the engine exhaust manifold to cause further combustion of unburned gases and to create an oxidizing atmosphere for the catalyst.” 17 It was this system to which the data base was initially narrowed: only cars using this kind of system were to be considered in making the “available technology” determination.

The problem the Administrator faced in making a determination that technology was available, on the basis of these data, was that actual tests showed only one car with actual emissions which conformed to the standard prescribing a maximum of .41 grams, per mile, of HC and 3.4 grams per mile of CO.18 No car had actually been driven 50,000 miles, the statutory “useful life” of a vehicle and the time period for which conformity to the emission standards is required.19 In the view of the EPA Administrator, however, the reasons for the high test readings were uncertain or ambivalent.

Instead, certain data of the auto companies were used as a starting point for making a prediction, but remolded into a more useable form for this purpose. As the Administrator put it :20

Much of the data reports emissions measured by test procedures different from the 1975 Federal test procedure and requires conversion to the 1975 procedure by calculations which cannot be regarded as precise. Emission data was frequently submitted without an adequate description of the vehicle being tested, the emission control systems employed, or the- purpose of the test. The fuel and oil used in tests were not always specified. Adjustments made to components of the engine or emission control system were frequently made and seldom'fully explained. In most cases, tests were not repeated, even where results departed significantly from established trends, and little or no information was submitted to explain the diagnosis of failure, where test results showed poor results. Most important, only a few test cars were driven to 20,000 miles or more, and no vehicle employing all components of any applicant’s proposed 1975 control systems has yet been driven to 50,000 miles. In the face of these difficulties, analysis and interpretation of the data required assumptions and analytical approaches which will necessarily he controversial to some degree, (emphasis added)

In light of these difficulties, the Administrator “adjusted” the data of the auto companies by use of several critical assumptions.

First, he made an adjustment to reflect the assumption that fuel used in 1975 model year cars would either contain an average of .03 grams per gallon or .05 grams per gallon of lead.21 This usually resulted in an increase of emissions predicted, since many companies *626had tested their vehicles on lead-free gasoline.

Second, the Administrator found that the attempt of some companies to reduce emissions of nitrogen oxides below the 1975 Federal standard of 3.0 grams per vehicle mile 22 resulted in increased emissions of hydrocarbons and carbon monoxide. This adjustment resulted in a downward adjustment of observed HC and CO data, by a specified factor 23

Third, the Administrator took into account the effect the “durability” of the preferred systems would have on the emission control obtainable. This required that observed readings at one point of usage be increased by a deterioration factor (DF) to project emissions at a later moment of use. The critical methodological choice was to make this adjustment from a base of emissions observed at 4000 miles. Thus, even if a car had actually been tested over 4000 miles, predicted emissions at 50,000 miles would be determined by multiplying 4000 mile emissions by the DF factor.24

Fourth, the Administrator adjusted for “prototype-to-production slippage.” This was an upward adjustment made necessary by the possibility that prototype cars might have features which reduced HC and CO emissions, but were not capable of being used in actual production vehicles 25

Finally, in accord with a regulation assumed, as to substance, in the text of the Decision, but proposed after the suspension hearing,26 a .downward adjustment in the data readings was made on the basis of the manufacturers’ ability, in conformance with certification procedures, to replace the catalytic converter “once during 50,000 miles of vehicle operation,” a change they had not used in their testing.27

With the data submitted and the above assumptions, the Administrator concluded that no showing had been made that requisite technology was not available. The EPA noted that this did not mean that the variety of vehicles produced in 1975 would be as extensive as before. According to EPA, “Congress clearly intended to require major changes in the kinds of automobiles produced for sale in the United States after 1974” and there “is no basis, therefore, for construing the Act to authorizing suspension of the standards simply because the range of performance of cars with effective emission control may be restricted as compared to present cars.” As long as “basic demand” for new light duty motor vehicles was satisfied, the applicants could not establish that technology was not available 28

For purposes of judicial review, the initial EPA decision rests on the technology determination. The Administrator did state:29

On the record before me, I do not believe that it is in the public interest to grant these applications, where compliance with 1975 standards by application of present technology can probably be achieved, and where ample additional time is available to manufacturers to apply existing technology to 1975 vehicles. (Emphasis added.)

The statute apparently contemplates the possibility of an EPA denial of suspension for failure to meet criterion (i) of § 202(b)(5)(D) (“essential to the public interest”) even though criterion (iii) has been satisfied (“applicant has established that effective control technology . [is] not available”).30 It suffices here *627to say that the EPA’s 1972 “public interest” finding was obviously only a restatement of, and dependent on the validity of, the conclusion of a failure to satisfy standard (iii) by showing that effective control technology is not available.

The Administrator also offered some “comments” on issues pertinent to the required “good faith” determination under standard (ii), as guidance to applications who might seek a one year suspension next year of the 1976 oxides of nitrogen standard. But he explictly disclaimed reaching that question in this proceeding. The thrust of his comment was to call into question the rigid “arms length” relationship structure which vehicle manufacturers imposed on their suppliers, as a source of a halter on progress in developing the required technology.31

C. This Court’s December 1972 Remand

After oral argument to this court on December 18, 1972, in a per curiam order issued December 19, 1972, we remanded the record to the Administrator, directing him to supplement his May 12, 1972 decision by setting forth:

(a) the consideration given by the Administrator to the January 1, 1972 Semiannual Report on Technological Feasibility of the National Academy of Sciences; and (b) the basis for his disagreement, if any, with the findings and conclusion in that study concerning the availability of effective technology to achieve compliance with the 1975 model year standards set forth in the Act.

Our remand order was not intended to indicate that we had concluded that an EPA conclusion was required as to clause (iv> — concerning the evaluation based on the NAS study and other information (from sources other than applicants)— when the Administrator had determined under (iii) that the auto companies had not shown technology was not available. We were nevertheless troubled by arguments advanced by petitioners that the methodology used by the Administrator in reaching his conclusion, and indeed the conclusion itself, was inconsistent with that of the Academy. It was our view that if and to the extent such differences existed they should be explained by EPA, in order to aid us in determining whether the Administrator’s conclusion under (iii) rested on a reasoned basis.

D. Supplement to the Decision of the Administrator

Our remand of the record resulted in a “Supplement to Decision of the Administrator” issued December 30, 1972. The Administrator in his Supplement stated that “In general I consider the factual findings and technical conclusions set forth in the NAS report and in the subsequent Interim Standards Report dated April 26, 1972 . to be consistent with my decision of May 12, 1972.” 32

The Report made by the NAS, pursuant to its obligation under 202(b)(5) (D) of the Clean Air Act, had concluded : “The Committee finds that the technology necessary to meet the requirements of the Clean Air Act Amendments for 1975 model year light-duty motor vehicles is not available at this time.” 33

The Administrator apparently relied, however, on the NAS Report to bolster his conclusion that the applicants had not established that technology was un*628available. The same NAS Report had stated:34

the status of development and rate of progress made it possible that the larger manufacturers will be able to produce vehicles that will qualify, provided that provisions are made for catalyst replacement and other maintenance, for averaging emissions of production vehicles, and for the general availability of fuel containing suitably low levels of catalyst poisons.

The Administrator pointed out that two of NAS’s provisos — catalytic converter replacement and low lead levels — had been accounted for in his analysis of the auto company data, and provision therefor had been insured through regulation.35 As to the third, “averaging emissions of production vehicles,” 36 the Administrator offered two reasons for declining to make a judgment about this matter: (1) The significance of averaging related to possible assembly-line tests, as distinct from certification test procedure, and such tests had not yet been worked out. (2) If there were an appropriate assembly-line test it would be expected that each car’s emissions could be in conformity, without a need for averaging, since the assembly line vehicles “equipped with fresh catalysts can be expected to have substantially lower emissions at zero miles than at 4000 miles.”37

The Administrator also claimed that he had employed the same methodology as the NAS used in its Interim Standards Report, evidently referring to the use of 4000 mile emissions as a base point, and correction for a deteriorrtion factor and a prototype-production slippage factor.38 The identity of methodology was also indicated, in his view, by the fact the EPA and NAS both agreed on the component parts of the most effective emission control system.

The Administrator did refer to the “severe driveability problems” underscored by the NAS Report, which in the judgment of NAS “could have significant safety implications,” 39 stating that he had not been presented with any evidence of “specific safety hazard” nor knew of any presented to the NAS. He did not address himself to the issue of performance problems falling short of specific safety hazards.

II. REJECTION OF MANUFACTURERS’ GENERAL CONTENTIONS

We begin with consideration, and rejection, of the broad objections leveled by petitioners against EPA’s over-all approach.

A. Future Technological Developments

We cannot accept petitioners’ arguments that the Administrator’s determination whether technology was “available,” within the meaning of section 202(b) (5) (D) of the Act, must be based solely on technology in being as of the time of the application, and that the requirement that this be “available” precludes any consideration by the Administrator of what he determines to be the “probable” or likely sequence of the technology already experienced. Congress recognized that approximately two years’ time was required before the start of production for a given model year, for the preparation of tooling and manufacturing processes.40 But Congress did not decide — and there is no reason *629for us to do so — -that all development had to be completed before the tooling-up period began. The manufacturers’ engineers have admitted that technological improvements can continue during the two years prior to production.41 Thus there was a sound basis for the Administrator’s conclusion that the manufacturers could “improve, test, and apply” technology during the lead time period.42 ■ The petitioners’ references to the legislative history are unconvincing. None of the statements quoted in their briefs specifically states that “available” as used in the statute means “available in 1972.” There is even comment that points to a contrary interpretation.43 In any event, we think the legislative history is consistent with the EPA’s basic approach and evidences no ascertainable legislative intent to the contrary.

While we reject the contention as broadly stated, principally by General Motors, we hasten to add that the Administrator’s latitude for projection is subject to the restraints of reasonableness, and does not open the door to “ ‘crystal ball’ inquiry.”44 The Administrator’s latitude for projection is unquestionably limited by relevant considerations of lead time needed for production.45 Implicit also is a requirement of reason in the reliability of the EPA projection. In the present case, the Administrator’s prediction of available technology was based on known elements of existing catalytic converter systems. This was a permissible approach subject, of course, to the requirement that any technological developments or refinements of existing systems, used as part of the EPA methodology, would have to rest on a reasoned basis.

B. Claimed Bight of Cross-Examination

Chrysler has advanced a due process claim based upon two principal features of the proceeding, the inability to engage in cross-examination and the inability to present arguments against the methodology used in the Technical Appendix of the Administrator, which served as a basis for his decision.

The suspension provision of Section 202(b)(5)(D) does not require a trial type hearing. It provides:

Within- 60 days after receipt of the application for any such suspension, and after public hearing, the Administrator shall issue a decision granting or refusing such suspension.

First, this provision for a “public hearing” contrasts significantly with other provisions that specifically require an adjudicatory hearing.46 More importantly, the nonadjudicatory nature of the “public hearing” contemplated is underscored by the 60 day limit for a decision to be made. The procedure contemplated by Congress in its 1970 legislation must be appraised in light of its *630concern with “avoidance of previous cumbersome and time-consuming procedures,” see Kennecott Copper Corp. v. EPA, 149 U.S.App.D.C. 231, 234, 462 F.2d 846, 849 (1972).

As to legislative history of this provision, the starting point is the provision in Senate Bill 4358:47

Upon receipt of such application, the Secretary shall promptly hold a public bearing to enable such manufacturer or manufacturers to present information relevant to the implementation of such standard. The Secretary, in his discretion, may permit any interested person to intervene to present information relevant to the implementation of such standard.

This was dropped in conference, along with a provision permitting six months for a suspension decision. The resulting legislation both expedited the decision-making, and contemplated EPA solicitation of a wide range of views, from sources other than the auto companies, though the companies’ applications and presentation would surely be the focus of consideration. Underlying this approach of both shortening time for decision and enlarging input lies, we think, an assumption of an informative but efficient procedure without mandate for oral cross examination.

In context, the “public hearing” provision amounts to an assurance by Congress that the issues would not be disposed of merely on written comments, the minimum protection assured by the Administrative Procedure Act for rule-making, but would also comprehend oral submissions of a legislative nature. These are required even for rule-making when “controversial regulations governing competitive practices” are involved. American Airlines, Inc. v. CAB, 123 U.S.App.D.C. 310, 317, 359 F.2d 624, 631 (en banc 1966), cert. denied, 385 U.S. 843, 87 S.Ct. 73, 17 L.Ed.2d 75 (1966); Walter Holm & Co. v. Hardin, 145 U.S.App.D.C. 347, 449 F.2d 1009 (1971). Even assuming oral submission, in a situation where “general policy” is the focal question, a legislative-type hearing is appropriate.48

A complication is presented by the case before us in that the general policy questions became interfused with relatively specific technical issues. Yet within the context of a quasi-legislative hearing and the time constraints of the statute, we do not think the absence of a general right of cross-examination on the part of the companies was a departure from “basic considerations of fairness.” Walter Holm & Co. v. Hardin, supra, 145 U.S.App.D.C. at 354, 449 F.2d at 1016. Hearings ran for two weeks and a wide range of participants was included within the proceeding: manufacturers, vendors of the control devices and public interest groups. The auto companies were allowed to submit written questions to the Hearing Panel to be asked to various witnesses. Opportunity to prepare written questions is not as satisfactory to counsel as the opportunity to proceed on oral cross-examination, with questions that develop from previous answers. But examination on interrogatories has long been used in the law when necessary, albeit second best. And *631interrogatories to a live witness — often arranged in private lawsuits by use of a commission — avoid the peril of “canned” affidavits and counsel-assisted, or even counsel-drafted, responses to interrogatories. Their availability was a reasonable attempt by EPA to elicit the facts and at the same time cope with the time constraints. We do not think more was required. There was a meaningful opportunity to be heard. The specific nature of a “hearing” varies with circumstances. Cafeteria & Restaurant Workers Union v. McElroy, 367 U.S. 886, 895, 81 S.Ct. 1743, 6 L.Ed.2d 1230 (1961), cited with approval in Goldberg v. Kelly, 397 U.S. 254, 263, 90 S.Ct. 1011, 25 L.Ed.2d 287 (1970). Whether particular attributes of forensic presentation are not only salutary but also mandatory must also depend on circumstances. The heft of the hearing problem, including the time constraints on decisions, convinces us that the assertion of a broad right of cross-examination cannot be successfully maintained.

We distinguish between the assertion of a. broad right of cross-examination, such as that argued to this court, and a claim of a need for cross-examination of live witnesses on a subject of critical importance which could not be adequately ventilated under the general procedures. This is the kind of distinction that this court made in its en bane opinion in American Airlines v. CAB, supra, 123 U.S.App.D.C. at 318-319, 359 F.2d at 632-633. We see no principled manner in which firm time limits can be scheduled for cross-examination consistent with its unique potential as an “engine of truth” — the capacity given a diligent and resourceful counsel to expose subdued premises, to pursue evasive witnesses, to “explore” the whole witness, often traveling unexpected avenues.

Given the variances in counsel, the reality that seasoning and experience are required even for trial judges who seek to avoid repetitive and undue cross-examination, the enhancement of difficulties encountered with the breadth of issues involved in a “public interest” proceeding, the fairly-anticipated problem of provision for redirect (and recross) and the interplay of different cross-examinations, there is not insignificant potential for havoc. What' is most significant is that these complications are likely to be disproportionate to the values achieved, in a proceeding focusing on technical matters where other techniques generally are sufficient to adduce the pertinent information- as to both what is known and unknown.

In context, we consider that the technique, adopted by EPA, of pre-screen-ing written questions submitted in advance is reasonable and comports with basic fairness as the general procedure. This approach permits screening by the hearing officer so as to avoid irrelevance and repetition, permits a reasonable estimate of the time required for the questioning, and aids scheduling and allocation of available time among various participants and interests.49 The record reveals that the hearing officers did not propound the pre-submitted questions like robots; they were charged with conducting a hearing for the purpose of focusing information needed for decision, and they quite appropriately “followed up” on questions.

We revert to our observation that a right of cross-examination, consistent with time limitations, might well extend to particular eases of need, on critical points where the general procedure proved inadequate to probe “soft” and sensitive subjects and witnesses. No such circumscribed and justified requests were made in this proceeding.

C. Right To Comment on EPA Methodology

A more serious problem, at least from the point of an informed decision-making process, is posed by the *632inability of petitioners to challenge the methodology of EPA at the hearing. In other contexts, it is commonplace for administrative proceedings to focus in detail on agency methodology,50 and such elucidation is salutary, of particular aid to a reviewing court. Again, however, we cannot ignore the problem of time. In part, EPA developed its methodology on the basis of submissions made by the companies at the hearings, as to the parameters of its various data. The requirement of submission of a proposed rule for comment does not automatically generate a new opportunity for comment merely because the rule promulgated by the agency differs from the rule it proposed, partly at least in response to submissions.51 Given the circumstances, we cannot hold the absence of the right to comment on the methodology a violation of the statute or due process, though such opportunity would certainly have been salutary.

While the statute makes no express provision therefor, we assume that Congress contemplated a flexibility in the administrative process permitting the manufacturers to present to EPA any comments as to its methodology, in a petition for reconsideration or modification. However, this opportunity does not permit invocation of the doctrine of failure to exhaust administrative remedies as a bar to these appeals, for those petitions could not have affected or deferred the finality of the EPA decision or the time for seeking judicial review. The opportunity is noted to obviate any possibility that the law, or our comments, may be misunderstood to require a rigid procedure of prompt and unshakeable decision-making. Our own December remand requesting clarification of the Decision illustrates that while this statute imposes some unusual time restraints it does not jettison the flexibility and capacity of reexamination that is rooted in the administrative process. American Airlines v. CAB, supra, 123 U.S.App.D.C. at 319; 359 F.2d at 633.

As matters have shaped up, the central technical issue on this appeal concerns the reliability of EPA’s methodology. While we do not say that the failure to provide reasonable opportunity to comment on EPA methodology invalidates the EPA Decision for lack of procedural due process, or similar contention, we must in all candor accompany that ruling with- the comment that the lack of such opportunity has had serious implications for the court given the role of judicial review.

We shall subsequently develop the legal questions, primarily questions of EPA’s burden of proof, that arise with respect to EPA methodology. We preface these with admission of our doubts and diffidence. We are beset with contentions of petitioners that bear indicia of substantiality. Yet we have no EPA comment on the specific questions raised, apart from some discussion by counsel which is not an adequate or appropriate substitute.52 Our December 1972 re*633mand opened the door to a candid discussion of these matters, but EPA fashioned a carefully limited response.

The EPA might have indicated that it desired to take a fresh look at its methodology on the basis of petitioners’ criticisms, in which case, on an adaptation of the Smith v. Pollin,53 procedure, this court might have remanded the ease to the agency. This remand would come during the course of our judicial review and would not conflict with the 60-day statutory time limit for the hearing and decision on the applications for suspension.

Indeed, the fact that the Administrator issued the Technical Appendix almost three months after his Decision, at a time when judicial review had already begun to run its course, indicates that the agency did not believe that agency consideration was frozen from the moment that the suspension decision was rendered, a view we approve. The EPA had latitude to continue further consideration even without requesting a court remand (under Smith v. Pollin) that would suspend judicial consideration.

III. OVERALL PERSPECTIVE OF SUSPENSION ISSUE

This case ultimately involves difficult issues of statutory interpretation, as to the showing required for applicants to sustain their burden that technology is not available. It also taxes our ability to understand and evaluate technical issues upon which that showing, however it is to be defined, must rest. At the same time, however, larger questions are at stake. As Senator Baker put it, “This may be the biggest industrial judgment that has been made in the United States in this century.” 116 Cong.Rec. 33,085 (1970). This task of reviewing the suspension decision was not assigned to us lightly. . It was the judgment of Congress that this court, isolated as it is from political pressures, and able to partake of calm and judicious reflection would be a more suitable forum for review than even the Congress.54

Two principal considerations compete for our attention. On the one hand, if suspension is not granted, and the prediction of the EPA Administrator that effective technology will be available is proven incorrect, grave economic consequences could ensue. This is the problem Senator Griffin described as the “dangerous game of economic roulette.” 116 Cong.Rec. 33,081 (1970). On the other hand, if suspension is granted, and it later be shown that the Administrator’s prediction of feasibility was achievable in 1975 there may be irretrievable ecological costs. It is to this second possibility we first turn.

A. Potential Environmental Costs

The most authoritative estimate in the record of the ecological costs of a one-year suspension is that of the NAS Report. Taking into account such “factors as the vehicle-age distribution among all automobiles, the decrease in vehicle miles driven per year, per car as vehicle age increases, the predicted nationwide growth in vehicle miles driven each year” and the effect of emission standards on exhaust control, NAS concluded that:55

. the effect on total emissions of a one-year suspension with no additional interim standards appears to be small. The effect is not more significant because the emission reduction now required of model year 1974 vehicles, as compared with uncontrolled vehicles (80 percent for HC and 69 percent for CO), is already so substantial.

*634Other considerations may diminish the costs even further. There seems to be agreement that there are performance costs for automobiles in employing pollution control devices, even if the effects on performance cannot fairly be characterized as constituting safety hazards. The NAS Report summarized the problem, as follows:56

Three areas of vehicle performance are likely to be adversely affected by the 1975 emission control systems. These are fuel economy, vehicle-acceleration capability, and vehicle drive-ability (or ability to perform adequately in all normal operating modes and ambient conditions).

The question in this context is not whether these are costs the consumer' should rightly bear if ecological damage is to be minimized, but rather the general effect on consumer purchasing of 1975 model year cars in anticipation of lower performance. A drop-off in purchase of 1975 ears will result in a prolonged usage of older cars with less efficient pollution control devices. If the adverse performance effect deterred purchasing significantly enough, resulting in greater retention of “older” ears in the “mix” of cars in use, it might even come to pass that total actual emissions (of all cars in use) would be greater under the 1975 than the 1974 standards.

Many of the anticipated performance problems are traceable to the systems introduced to conform cars to control of nitrogen oxides to achieve prescribed 1975 standards, by use of exhaust-gas recycle (EGR). Such systems affect vehicle-acceleration capability because the power output for a given engine displacement, engine speed, and throttle setting is reduced.57 The NAS Report indicates that such systems could result in direct fuel-economy penalties of up to 12 percent compared with 1973 prototype vehicles.58

The NAS Report states that the effects of emission controls on vehicle driveability are difficult to quantify, but nevertheless makes the following qualitative evaluation:59

Driveability after a cold-engine start, and especially with cold ambient conditions, is likely to be impaired. To reduce HC and CO emissions during engine warmup, the choke is set to release quickly, and the fuel-air mixture is leaned out as early as possible after engine startup. Under these conditions, problems of engine stall, and vehicle stumble and hesitation on rapid acceleration, have been prevalent.

The willingness of the consumer to buy 1975 model year cars may also be affected, to some degree, by the anticipated significant costs of pollution control devices. The problem is further bedeviled by the possibility that consumers albeit rightly assigned the cost burden of pollution devices, may seek to avoid that burden, however modest,60 and to exercise, at least in some measure, an option to use older cars. Again, this would have the thrust of increasing actual total emissions of cars in use.

We may also note that it is the belief of many experts — both in and out of the automobile industry — that air pollution cannot be effectively checked until the industry finds a substitute for the conventional automotive power plant — the reciprocating internal combustion (i. e., *635“piston”) engine.61 According to this view, the conventional unit is a “dirty” engine. While emissions from such a motor can be “cleaned” by various thermal and catalytic converter devices, these devices do nothing to decrease the production of emissions in the engine’s combustion chambers. The automobile industry has a multi-billion-dollar investment in the conventional engine, and it has been reluctant to introduce new power plants or undertake major modifications of the conventional one.62 Thus the bulk of the industry’s work on emission control has focussed narrowly on converter devices. It is clear from the legislative history that Congress expected the Clean Air Amendments to force the industry to broaden the scope of its research — to study new types of engines and new control systems.63 Perhaps even a one-year suspension does not give the industry sufficient time to develop a new approach to emission control and still meet the absolute deadline of 1976. If so, there will be ample time for the EPA and Congress, between now and 1976 to reflect on changing the statutory approach. This kind of cooperation, a unique three-way partnership between the legislature, executive and judiciary, was contemplated by the Congress 64 and is apparent in the provisions of the Act.65

The NAS estimated that there would be a small environmental cost to suspension of 1975 standards even if 1974 standards were retained, but further recommended intermediate standards that would dilute even such modest environmental cost66 The following table shows the various standards, and one put forward by Ford for 1975:

Maximum emissions (grams per mile)

HC CO

1974 standards ............ 3.4 39.0

Ford proposal ............. 1.6 19.0

NAS recommendation for

Intermediate standards:

No catalyst change..... 1.1 8.2

One catalyst change .... 0.8 6.3

1975 Standards............41 3.4

Our concern that the 1975 standards may possibly be counter-productive, due to decreased driveability and increased cost, is not to be extrapolated into a caution against any improvement, and concomitant reduction in permitted emissions. In such matters, as the NAS recommendation for interim standards implicitly suggests, a difference in degree may be critical, and the insistence on absolute 1975 standards, without suspension or intermediate level, may stretch for the increment that is essentially counter-productive.

*636We also observe that Ford Motor Company is on record as to capability of greater emission controls, i. e., lower level of emissions, than those permitted for 1974 model year cars,67 and Ford proposed that, given certain regulatory assumptions,68 the Administrator adopt an interim standard of 1.6 gm/mi HC and 19.0 gm/mi CO levels, about one half those permitted for the 1974 model year cars.

On balance the record indicates the environmental costs of a one-year suspension are likely to be relatively modest. This must be balanced against the potential economic costs — and ecological costs — if the Administrator’s prediction on the availability of effective technology is incorrect.

B. Potential Economic Costs

Theoretical possibility of industry shutdown

If in 1974, when model year 1975 cars start to come off the production line, the automobiles of Ford, General Motors and Chrysler cannot meet the 1975 standards and do not qualify for certification, the Administrator of EPA has the theoretical authority, under the Clean Air Act, to shut down the auto industry, as was clearly recognized in Congressional debate.69 We cannot put blinders on the facts before us so as to omit awareness of the reality that this authority would undoubtedly never be exercised, in light of the fact that approximately 1 out of every 7 jobs in this country is dependent on the production of the automobile.70 Senator Muskie, the principal sponsor of the bill, stated quite clearly in the debate on the Act that he envisioned the Congress acting if an auto industry shutdown were in sight.71

The economic consequence of an approach geared to stringency, relying on relaxation as a safety valve

A more likely forecast, and one which enlightens what influenced the EPA decision to deny the suspension, was articulated by George Allen, Deputy Assistant Administrator for General Enforcement and a member of EPA’s Hearing Panel:72

The problem really comes down to this: A decision has to be made next month, early next month. If the decision is to suspend the standards and adopt an interim standard . and in 1975 it turns out that technology exists to meet the statutory standard, today’s decision turns out to be wrong.

* * * * * #

If, on the other hand, a decision is made today that the standards cannot lawfully be suspended, and we go down to 1975 and nobody can meet the standard, today’s decision was wrong.

In [the first] case, there is not much to do about the wrong decision; it was made, many people relied on it; it turns out the standard could have been met, but I doubt if we could change it.

In the second case, if a wrong decision is made, there is probably a remedy, a re-application and a recognition by the agency that it is not technically feasible to meet the standards. You can correct the one; you probably can’t correct the other.

*637Grave problems are presented by the assumption that if technical feasibility proves to be a “wrong decision” it can be remedied by a relaxation.

Certain techniques available to the Administrator, through changes in the certification procedure, can be used in an even handed manner for all three auto companies to facilitate compliance with the 1975 standards. Already lower lead levels in fuel available for 1975 model year cars have been prescribed to increase the efficiency of the catalytic converter. Similarly certain changes in the regulatory system, through allowable maintenance and permitted change in the catalytic converter, have been made by EPA. These techniques work with reasonable impartiality as to the various auto companies.

However, a relaxation of standards, and promulgation of an interim standard, at a later hour — after the base hour for “lead time” has been passed, and the production sequence set in motion — forebodes quite different consequences. The record before us suggests that there already exists a technological gap between Ford and General Motors,73 in Ford’s favor. General Motors did not make the decision to concentrate on what EPA found to be the most effective system at the time of its decision — the noble metal monolithic catalyst. Instead it relied principally on testing the base metal catalyst as its first choice system.74 In predicting that General Motors could meet the 1975 standards, EPA employed a unique methodological approach. Instead of taking emissions at 4000 miles of cars with preferred systems — with which none of the General Motors cars was equipped — and applying against this, adjustments for lead levels and deterioration, as had been done in the case of Ford and Chrysler, EPA took emissions at 4000 miles of GM cars which had no converters of any kind, and predicted how they would function with an Engelhard monolithic catalytic converter, based on auto manufacturers’ use of this device in a number of cars — principally Ford’s — when testing it for durability.75 In his Supplemental Decision the Administrator recognized that this was a departure from NAS methodology, stating :76

In its Interim Standards Report the National Academy recommended a methodology for predicting the emission levels achievable by manufacturers. This recommended methodology is the same methodology that was employed in the technical appendix to my decision in evaluating the test results of all manufacturers except General Motors. (Emphasis added.)

The case is haunted by the irony that what seems to be Ford’s technological lead 77 may operate to its grievous detriment, assuming the relaxation-if-necessary approach voiced by Mr. Allen,78 If *638in 1974, when certification of production vehicles begins, any one of the three major companies cannot meet the 1975 standards, it is a likelihood that standards will be set to permit the higher level of emission control achievable by the laggard. This will be the case whether or not the leader has or has not achieved compliance with the 1975 standards. Even if the relaxation is later made industry-wide, the Government’s action, in first imposing a standard not generally achievable and then relaxing it, is likely to be detrimental to the leader who has tooled up to meet a higher standard than will ultimately be required.

In some contexts high achievement bestows the advantage that rightly belongs to the leader, of high quality. In this context before us, however, the high achievement in emission control results, under systems presently available, in lessened car performance — an inverse correlation. The competitive disadvantage to the ecological leader presents a forbidding outcome — if the initial assumption of feasibility is not validated, and there is subsequent relaxation — for which we see no remedy.79

C. Light Weight Trucks

We now take up the serious contention of International Harvester (IH) that the EPA decision effectively rules out the production of 1975 model year IH light weight trucks and multi-purpose passenger vehicles (MPVs). This requires us to focus on the Administrator’s conception that the 1970 Clean Air Act envisioned restricting production of vehicles to that necessary to fill “basic demand.” 80

The Administrator does not dispute International Harvester’s claim that it will not be able to produce the vehicles in question, and indeed the limited testing of one of its MPVs showed, even as evaluated by EPA methodology, that such standards could not be achieved.81 Yet a suspension was not granted, presumably for the reasons advanced by EPA to this court, that International Harvester was “required to alter the performance characteristics of its vehicles in the interest of meeting the 1975 emission standards.” 82 The inability of IH vehicles to meet the standards seems accountable by the uses to which they are *639put, hauling large loads or towing heavy trailers. To serve this purpose vehicles must be designed with higher than normal axle ratios, thus requiring greater power from the engine and producing higher exhaust gas temperatures in order to attain any given speed.83 Therefore, for all practical purposes a redesign of performance characteristics will preclude the present uses to which IH vehicles are put.

The Administrator, nonetheless, takes the position that International Harvester can be denied a suspension because he has found that “new car demand” will be satisfied by the production of the major auto companies, and thus apparently posits that the absence from the 1975 market of all light weight trucks and MPVs is fully consistent with the Act. We cannot agree.

Section 202(b)(1) of the Act applies its drastic standards to 1975 models of “light duty vehicles.” It is our view that the legislative history reveals this term to mean “passenger cars.” In the Report of the Senate Committee on Public Works on S.4358,84 the Committee clearly distinguished between the automobile, which must “meet a rigid timetable and a high degree of emission control compliance,” and other vehicles, such as “trucks and buses and other commercial vehicles,” which are governed by a different authority to promulgate standards. At another point of the Senate Report, the legislative use of the term light duty vehicles, as interchangeable with passenger cars, is made even more clear;85

The authority provided in section 202 (a) would continue to be available to the [Administrator] to establish standards for light duty motor vehicles (passenger cars) during the period prior to and following the effective date of the standards established by subsection (b).

References abound in Congressional debate to the same effect.86 This kind of legislative intent must be given priority, in interpreting this law, over any presumption of continuance of prior administrative definitions of this term87 or to the policy of upholding reasonable interpretations of statutes by administrative agencies 88 in the absence of other discernible legislative intent. Volkswagenwerk v. FMC, 390 U.S. 261, 272, 88 S.Ct. 929, 19 L.Ed.2d 1090 (1967); Greater Boston Television Corp. v. FCC (I), 143 U.S.App.D.C. 383, 392, 444 F.2d 841, 850, cert. denied, 403 U.S. 923, 91 S.Ct. 2229, 29 L.Ed.2d 701 (1971).

For the above reasons we cannot sustain the definition of “Light duty vehicle” as: 89

any motor vehicle either designed primarily for transportation of property and rated at 6,000 pounds GVW or less or designed primarily for transportation of persons and having a capacity of 12 persons or less

to the extent that it includes light weight trucks in the category that must meet the drastic emission reduction standards *640set for 1975 models. These light weight trucks will be governed by the standards duly promulgated by EPA for “trucks and buses and other commercial vehicles.”

This is not to say that the modification of the “light duty vehicles” definition must exclude MPVs, which largely overlap in their usage with passenger cars. We merely hold the present regulation contrary to legislative intent. We have jurisdiction to decide this issue, even though the reasonableness of the regulation could be challenged in a separate proceeding in the District Court,90 because the validity of the regulation is a premise of the refusal to grant suspension. “It would be an empty and useless thing to review an order . . . based on a regulation the validity of which might be subsequently nullified.” Doe v. Civil Aeronautics Board, 356 F.2d 699, 701 (10th Cir. 1966).

We decline the proposal of International Harvester, therefore, that only its vehicles be granted a suspension. Light weight trucks of other manufacturers, such as Ford, equally demonstrated an inability to comply with the 1975 standards.91 Under the view taken here, the light weight trucks of all manufacturers are properly exempted from the scope of “light duty vehicles.” This comports with competitive as well as statutory considerations, as the Administrator’s own brief delineates: 92

If International Harvester is granted a suspension, it should be able to sell its vehicles at a lower cost than competitors who met the standards. This is so because International Harvester’s 1975 models would not include expensive catalytic devices to control emissions. Also the Company’s vehicles would probably perform better for the same reason. Thus, if suspension is granted, it is likely that International Harvester will gain a substantial competitive advantage over manufacturers who sacrificed the performance of their vehicles, and perhaps profits, in order to comply with the 1975 standards.

Assuming light duty vehicles are defined by EPA to include MPVs a question may arise whether they are entitled-to a one-year suspension, for lack of feasibility, even though passenger vehicles generally should be denied a suspension. We shall not consider this question unless and until EPA has had an opportunity to address itself to the problems in the light of our opinion herein.

D. The Issue of Feasibility Sufficient for Basic Auto Demand

The foregoing conclusion is not to be misunderstood as amounting to an acceptance of another “basic demand” contention raised by the auto manufacturers. We are inclined to agree with the Administrator that as long as feasible technology permits the demand for new passenger automobiles to be generally met, the basic requirements of the Act would be satisfied, even though this might occasion fewer models and a more limited choice of engine types. The driving preferences of hot rodders are not to outweigh the goal of a clean environment.

A difficult problem is posed by the companies’ contention that the production and major retooling capacity does not exist to shift production from a large number of previous models and engine types to those capable of complying with the 1975 standards and meeting the demand for new cars. The Administrator made no finding as to this problem. We believe the statute requires such a finding, explaining how the Administrator estimates “basic demand” and how his definition conforms to the statutory objective. The emission standards set for 1976 cannot be breached, since they *641represent an absolute judgment of Congress. But as to the decision on a one-year suspension, and the underlying issue of technological feasibility, Congress intended, we think, that the Administrator should take into account such “demand” considerations.

A significant decrease in auto production will have a major economic impact on labor and suppliers to the companies. We have no reason to believe that “effective technology” did not comport within its meaning sufficient technology to meet a basic level of consumer demand.

E. Balancing of Risks

This case inevitably presents, to the court as to the Administrator, the need for a perspective on the suspension that is informed by an analysis which balances the costs of a “wrong decision” on feasibility against the gains of a correct one. These costs include the risks of grave maladjustments for the technological leader from the eleventh-hour grant of a suspension, and the impact on jobs and the economy from a decision which is only partially accurate, allowing companies to produce cars but at a significantly reduced level of output. Against this must be weighed the environmental savings from denial of suspension. The record indicates that these will be relatively modest. There is also the possibility that failure to grant a suspension may be counter-productive to the environment, if there is significant decline in performance characteristics.

Another consideration is present, that the real cost to granting a suspension arises from the symbolic compromise with the goal of a clean environment. We emphasize that our view of a one year suspension, and the intent of Congress as to a one year suspension, is in no sense to be taken as any support for further suspensions. This would plainly be contrary to the intent of Congress to set an absolute standard in 1976. On the contrary, we view the imperative of the Congressional requirement as to the significant improvement that must be wrought no later than 1976, as interrelated with the” provision for one-year suspension. The flexibility in the statute provided by the availability of a one-year suspension only strengthens the impact of the absolute standard. Considerations of fairness will support comprehensive and firm, even drastic, regulations, provided a “safety valve” is also provided — ordinarily a provision for waiver, exception or adjustment, in this case a provision for suspension.93 “The limited safety valve permits a more rigorous adherence to an effective regulation.” WAIT Radio v. FCC, supra, 135 U.S.App.D.C. at 323, 418 F.2d at 1159. To hold the safety valve too rigidly is to interfere with the relief that was contemplated as an integral part of the firmness of the overall, enduring program.

We approach the question of the burden of proof on the auto companies with the previous considerations before us.

IV. THE REQUIRED SHOWING ON “AVAILABLE TECHNOLOGY”

It is with utmost diffidence that we approach our assignment to review the Administrator’s decision on “available technology.” The legal issues are intermeshed with technical matters, and as yet judges have no scientific aides. Our diffidence is rooted in the underlying technical complexities, and remains even when we take into account that ours is a judicial review, and not a technical or policy redetermination, our review is channeled by a salutary restraint, and deference to the expertise of an agency that provides reasoned analysis. Nevertheless we must proceed to the task of judicial review assigned by Congress.

*642The Act makes suspension dependent on the Administrator’s determination that:

the applicant has established that effective control technology, processes, operating methods, or other alternatives are not available or have not been available for a sufficient period of time to achieve compliance prior to the effective data of such standards

A. Requirement of Observed Data From Manufacturers

Clearly this requires that the applicants come forward with data which showed that they could not comply with the contemplated standards. The normal rules place such a burden on the party in control of the relevant information.94 It was the auto companies who were in possession of the data about emission performance of their cars.

The submission of the auto companies unquestionably showed that no car had actually been driven 50,000 miles and achieved conformity of emissions to the 1975 standards. The Administrator’s position is that on the basis of the methodology outlined, "he can predict that the auto companies can meet the standards, and that the ability to make a prediction saying the companies can comply means that the petitioners have failed to sustain their burden of proof that they cannot comply.

B. Requisite Reliability of Methodology Relied on by EPA To Predict Feasibility Notwithstanding Lack of Actual Experience

We agree with the Administrator’s proposition in general. Its validity as applied to this case rests on the reliability of his prediction, and the nature of his assumptions. One must distinguish between prediction and prophecy. See EDF v. Ruckelshaus, 142 U.S.App.D.C. 74, 89, 439 F.2d 584, 597 (1971). In a matter of this importance, the predictor must make a showing of reliability of the methodology of prediction, when that is being relied on to overcome this “adverse” actual test data of the auto companies. The statute does not contemplate use of a “crystal ball.” See National Resources Defense Council, Inc. v. Morton, 148 U.S.App.D.C. 5, 15, 458 F.2d 827, 837 (1972).

The Administrator, however, raises a different issue by contending that the companies, wholly aside from his methodology, did not submit sufficient evidence to enable him to make the required determination as to “available technology.” This goes to the standard rather than the burden of proof, and comes close to adoption of “beyond a reasonable doubt” as the required showing. Aside from a possible finding of bad faith, which the Administrator specifically eschews making, this position cannot stand. The companies came forward with all the data that there was to be had, and the Administrator did not specifically ask for more. Additionally, our perspective on the interests furthered by a sound EPA decision, and jeopardized by a “wrong decision,” are material to the issue of standard of proof. This is a situation where, as we have stated, the risks of an erroneous denial of suspension outweigh the risks of an erroneous grant. On the issue of burden of proof, the standard adopted must take into account the nature and consequences of risk of error. See In re Winship, 397 U.S. 358, 371-372, 90 S.Ct. 1068, 25 L.Ed.2d 368 (1970) (Mr. Justice Harlan, concurring); U. S. v. Brown, 155 U.S.App.D.C.—, 478 F.2d 606 (1973). This view of the standard of proof dictates the standard normally adopted in civil matters, a preponderance of the evidence.95

*643Our approach relates considerations of ecological and economic costs, dealt with above, to the legal issue of burden and standard of proof. Nominally the statute, in § 202(b)(5)(D), sets forth separate criteria as to “public interest,” in clause (i), and “available technology,” in clause (iii). But the assignment of the burden and standard of proof on “available technology” inescapably involves many of the same considerations as those involved in a “public interest” determination, and it would have been helpful to this court if the Administrator had expressly commented on the public interest in this connection.

The underlying issue is the reasonableness and reliability of the Administrator’s methodology, for it alone offsets the data adduced by petitioners in support of suspension. It is the Administrator who must bear the burden on this matter, because the development and use of the methodology are attributable to his knowledge and expertise. When certain material “lies particularly within the knowledge” of a party he is ordinarily assigned the burden of adducing the pertinent information.96 This assignment of burden to a party is fully appropriate when the other party is confronted with the often-formidable task of establishing a “negative averment.” United States v. Denver & R.G.R. Co., 191 U.S. 84, 92, 24 S.Ct. 33, 48 L.Ed. 106 (1903). In the context of this proceeding, this requires that EPA bear a burden of adducing a reasoned presentation supporting the reliability of its methodology.

C. Analysis of EPA Assumptions

The multiple assumptions used by the Administrator in making his prediction are subject to serious doubts.

The basic formula used to make the prediction that each of the manufacturers could meet the 1975 standards was based on 1975 certification requirements, so that in part it paralleled testing procedures which would be used in 1975 to certify automobiles for sale. The formula is:97

Four kinds of assumptions were used in making the 50,000 mile emission prediction: (1) regulatory, (2) engineering or scientific, (3) techniques of application of basic formula to particular companies, and (4) statistical reliability of the final prediction.

1. Regulatory assumptions

First, EPA assumed that certain types of maintenance would have to be performed on 1975 model year cars, if its 50,000 miles emission predictions were to be meaningful. Subsequent to the issue of its Technical Appendix, a Proposed Rule Making formulated these requirements as part of 1975 certification procedure.98 This assumption was necessary because much of the data supplied by the companies was obtained from cars that were under rigid controls during testing.99 The problem with such maintenance assumptions is whether the ordinary driver will actually pay for this kind of maintenance just to reduce the *644emission levels of his automobile. It is one thing to build maintenance into the 1975 certification procedure, when fleet samples are durability tested. It is another to posit that such standards will be maintained, or are reasonably likely to be maintained, by consumers. A hard question is raised by the use of a methodological assumption without evidence that it will correspond to reality, or a reasonable and forthright prediction based on expertise.

Secondly, the predicted emission level assumes that there will be one total replacement of the catalytic converter at some time after 25,000 miles. This entered into the formula as an adjustment to the predicted deterioration factor.100 The critical question is how much will the one replacement reduce emissions otherwise obtainable by use of a single catalyst. This relationship had to be assumed because manufacturers had not used catalytic converter replacements in their testing. The Administrator admitted that this factor was imprecise.101 Yet, in the ease of General Motors, the use of the assumed value of this factor was critical in allowing the Administrator to make a 50,000 mile emission prediction under the 1975 standards.102

The third regulatory assumption relates to the average lead level which will exist in gasoline available for 1975 model year cars. Lead levels in gasoline contribute to the levels of HC and CO both in terms of normal emission control achievable (the 4000 mile emission) and to the deterioration in emissions over time (deterioration factor). Thus, in the case of the Chrysler car used to predict conformity with the 1975 standards, a .03 lead in gasoline produced 4000 mile emissions of .27 grams HC and 1.51 CO, whereas a .05 level of lead resulted in .29 and 1.66 grams respectively. Similarly .03 lead produced a corrected deterioration factor of .67 HC and 1.5 CO, whereas a .05 level produced .73 HC and 1.65 CO.103

On December 27, 1972, a regulation was promulgated “designed to assure general availability by July 1, 1974, of suitable gasolines containing no more than .05 grams per gallon of lead. . ” 104 It was the assumption of the Administrator that the .05 maximum would result in gas containing on the average .03 grams per gallon of lead. The discrepancy between the maximum and average is accounted for by the contamination of lead free gasoline from its point of production to its marketing outlet. Thus EPA will allow a maximum of .05 but anticipates that on the average fuel will be at .03. This assumption is, however, subject to testimony in the record indicating a difference between companies in their ability to achieve gasoline with a low lead level complying with the proposed regulation. Amoco said that its proposal for a .07 maximum “should result in effective lead levels of .02 to .03 grams of lead per gallon.” 105 Texaco did not think it could deliver gas to service stations at a lead level below .07.106 We cannot resolve whether a differential ability really exists, but we also have no refinement and resolution by the EPA (as distinguished from the briefs of its counsel). We do not say this matter is a critical defect; still it leaves a residue of uncertainty that beclouds the EPA assumption of a .03 average, needed in its methodology to predict conformity with the 1975 standards.

*6452. Engineering and scientific assumptions

Engineering or scientific assumptions are made in predicting 4000 mile emissions and deterioration factors, and we shall give separate consideration to each independent variable.

a. The 4000 mile emission factor

The use of 4000 mile emissions as a starting point is based on certification procedures.107 No challenge has been made to this mileage as a base point, largely because it appears that at this mileage the engine is broken in and emission levels are relatively stabilized.108 EPA decided to adjust raw data supplied, at least in the case of Ford and Chrysler, of emissions at 4000 miles to take account of a “Lead Adjustment Factor.” 109 This was done because in most cases emissions data reflected fuels with a close to zero lead level which had been used by the manufacturers in their testing programs.

Lead adjustment factor

This Lead Adjustment Factor was calculated using only Ford cars, but the value of the factor was assumed to be the same in adjusting Chrysler 4000 mile emissions with this factor.110 The cars had been tested with a dynamometer, a type of test equipment used for laboratory testing of an engine. A measurement of the efficiency of the catalytic converter at the 4000 mile mark was the critical value which had to be obtained from the dynamometer since this would indicate what the proper lead adjustment factor would be.111

EPA assumed that 200 hours on the dynamometer corresponded to 4000 miles usage, based on a critical and contested EPA assumption that the tests were conducted at 1000 RPM. Petitioners claim that the high temperature readings on the dynamometer reflect a higher RPM, and hence that a testing below 200 hours corresponded to 4000 miles of use. EPA disputes the steps in that chain of reasoning, and argues that a higher temperature may be attributable not to a RPM in excess of 1000, but to a heavy load on the vehicle, and in the alternative contends that even if there was a RPM greater than 1000, the speed may not have increased, due to a shift in gear.

The cause of higher than expected temperature readings cannot be ascertained from the record, and we are left with the alternative contentions of the parties. It is up to EPA, however, to support its methodology as reliable, and this requires more than reliance on the unknown, either by speculation, or mere shifting back of the burden of proof.112

*646b. Deterioration factor

Methodological problems also existed with the calculation of the deterioration factor, which took account of possible deterioration in emission quality from 4000 miles to 50,000 miles. Different questions arose as to the calculation of this factor for Ford and Chrysler.

In the case of Ford, the Administrator predicted that emissions would improve from 4000 to 50,000 miles, and arrived at a deterioration factor of less than l.113 He calculated average deterioration factors for Ford vehicles of .80 HC and .83 CO. This is to be compared with a deterioration factor of 2.5 used by NAS.114 The Administrator never explained why there should be no deterioration. Nor does EPA explain how this result can be squared with other data on Ford catalyst efficiencies, which was used in the case of the General Motors prediction, showing 50,000 mile catalyst efficiencies ranging from 21% to 53% for HC and 47% to 72% for CO.115

In the case of Chrysler, the deterioration factor was also calculated to be less than 1, but this figure was only arrived at after eliminating some data points from the emission measurement on the tested car #333, due to what EPA claimed were unrepresentative points resulting from non-catalyst malfunctions.116 Although it may be, as EPA argues here, that including the data points would still produce predicted 50,-000 emission levels in conformity with the 1975 standard, the fact remains that these data points were removed. Moreover, it is not apparent why one should ignore malfunctions of a car which contribute to high emissions, even if they are not malfunctions of the converter. Malfunctions of cars occur to some degree, and cars operating in 1975 will undoubtedly be subject to them.

Lead adjustment factor

A lead adjustment factor is applied to the deterioration factor, as well as to 4000 mile emissions. EPA estimated on the basis of the questionable Ford dynamometer data, that lead levels had no observable effect, which was contrary to industry testimony on the subject.117 The Administrator evidently had doubts as to the dependability of these results as well, and therefore assumed a 10% factor for lead adjustment.118 No explanation is given of the origins of this 10% figure. If the willingness to take some factor evidences distrust in the data, the question then becomes whether 10% is enough.

3. EPA methodology for General Motors

In the case of General Motors an entirely different methodology from that used for Ford and Chrysler was employed. This was adopted due to limited testing by GM of noble metal catalysts.

The methodology was to take the raw emission values produced by a GM car prior to catalyst treatment of any kind multiplied by a factor representing the efficiency of the catalyst, i. e., the percentage of a given pollutant that the catalyst converts to harmless vapor, in order to obtain the projected overall emission performance at 50,000 miles.119 These methods of" calculation were developed by the Administrator and were not used by NAS in their evaluation.120 *647The catalyst efficiency data were taken from Engelhard converters used principally on Ford cars and applied against the raw emissions of a General Motors engine. This assumed, with no explanation of the validity of such an assumption, that Engelhard catalysts will function as efficiently in General Motors cars as in those of Ford. A prediction was made on the basis of a hypothetical case. One cannot help be troubled by the adoption of this technique for General Motors. It was apparently recognized as at best a second best approach, in terms of the reliability of the prediction, or the same catalyst efficiency procedure would also have been used for Ford and Chrysler.

4. Statistical reliability of assumptions

In this case the Administrator is necessarily making a prediction. No tests exist on whether this prediction is or is not reliable. It would, therefore, seem incumbent on the Administrator to estimate the possible degree of error in his prediction. The NAS, for example, said that the data of the manufacturers were subject to ± 20-30% margin of error,121 and this is separate from any margin of error that may be due to the various assumptions made by the Administrator. It is not decisive to say, as EPA argues in its brief, that this is just a matter of quality control in production. The first issue is whether the automobile built with rigid adherence to specifications will perform as predicted. The issue of quality control, whether cars will indeed be built in accordance with specifications, raises a separate and additional problem.

The possibility of error must take into account that only 1 Ford car, 1 Chrysler car, and 1 hypothetical General Motors car form the foundation for predicted conformity with the 1975 standard.122 The Administrator would say that it is enough to validate the principle of the electric light bulb if only one is seen at work. But we do not yet have one that has worked; instead we have four predictions. Questions like these arise: (1) For how many different types of engines will these predictions be valid? (2) Does it make a difference that the tested cars were experimental and driven under the most controlled conditions? The best car analysis of EPA raises even further doubts when considered alongside the NAS Report which used 55 vehicles in arriving at its recommended interim standard.123

V. CONCLUSION AND DISPOSITION

We may sensibly begin our conclusion with a statement of diffidence.124 It is not without diffidence that a court undertakes to probe even partly into technical matters of the complexity of those covered in this opinion. It is with even moré diffidence that a court concludes that the law, as judicially construed, requires a different approach from that taken by an official or agency with technical expertise. Yet this is an inescapable aspect of the judicial condition, though we stay mindful of the overarching consideration that a court’s role on judicial review embraces that of a constructive cooperation with the agency involved in furtherance of the public interest.125

*648A court does not depart from its proper function when it undertakes a study of the record, hopefully perceptive, even as to the evidence on technical and specialized matters, for this enables the court to penetrate to the underlying decisions of the agency, to satisfy itself that the agency has exercised a reasoned discretion, with reasons that do not deviate from or ignore the ascertainable legislative intent.126

In this case technical issues permeate the “available technology” determination which the Administrator made the focal point of his decision. In approaching our judicial task we conclude that the requirement of a “reasoned decision” by the Environmental Protection Agency means, in present context, a reasoned presentation of the reliability of a prediction and methodology that is relied upon to overcome a conclusion, of lack of available technology, supported prima faciely by the only actual and observed data available, the manufacturers’ testing.

The number of unexplained assumptions used by the Administrator; the variance in methodology from that of the Report of the National Academy of Sciences, and the absence of an indication of the statistical reliability of the prediction, combine to generate grave doubts as to whether technology is available to meet the 1975 statutory standards. We say this, incidentally, without implying or intending any acceptance of petitioners’ substitute assumptions. These grave doubts have a legal consequence. This is customarily couched, by legal convention, in terms of “burden of proof.” We visualize the problem in less structured terms although the underlying considerations, relating to risk of error, are related. As we see it the issue must be viewed as one of legislative intent. And since there is neither express wording or legislative history on the precise issue, the intent must be imputed. The court must seek to discern and reconstruct what the legislature that enacted the statute would have contemplated for the court’s action if it could have been able to foresee the precise situation.127 It is in this perspective that we have not flinched from our discussion of the economic and ecological risks inherent in a “wrong decision” by the Administrator. We think the vehicle manufacturers established by a preponderance of the evidence, in the record before us, that technology was not available, within the meaning of the Act, when they adduced the tests on actual vehicles; that the Administrator’s reliance on technological methodology to offset the actual tests raised serious doubts and failed to meet the burden of proof which in our view was properly assignable to him, in the light of accepted legal doctrine and the intent of Congress discerned, in part, by taking into account that the risk of an “erroneous” denial of suspension outweighed the risk of an “erroneous” grant of suspension. We do not use the burden of proof in the conventional sense of civil trials, but the Administrator must sustain the burden of adducing a reasoned presentation supporting the reliability of EPA’s methodology.

EPA’s diligence in this proceeding, fraught with questions of statutory interpretation, technical difficulties and burdensome time constraints placed on the decision-making process, has been commendable. The agency was presented with a prickly task, but has acted expeditiously to carry out what it perceived to be a drastic mandate from Congress. This statute was, indeed, deliberately designed as “shock treatment” to the industry. Our central difference with the Administrator, simply put, stems from our view concerning the Congressional intent underlying the one year suspension pro*649vision. That was a purposeful cushion —with the twin purpose of providing “escape hatch” relief for 1975, and thus establishing a context supportive of the rigor and firmness of the basic standards slated for no later than 1976. In our view the overall legislative firmness does not necessarily require a “hard-nosed” approach to the application for suspension, as the Administrator apparently supposed, and may indeed be furthered by our more moderate view of the suspension issue, particularly in assigning to the Administrator the burden of producing a reasoned presentation of the reliability of his methodology. This is not a matter of clemency, but rather a benign approach that moderates the “shock treatment” so as to obviate excessive and unnecessary risk of harm.

Our decision is also responsive to the differences between the EPA decision and the NAS Report. Although in some instances “the factual findings and technical conclusions”128 are consistent with those of the Administrator, the NAS conclusion was that technology was not available to meet the standards in 1975. Congress called on NAS, with presumed reliance on the knowledge and objectivity of that prestigious body, to make an independent judgment. The statute makes the NAS conclusion a necessary but not sufficient condition of suspension. While in consideration of the other conditions of suspension, EPA was not necessarily bound by NAS’s approach, particularly as to matters interlaced with policy and legal aspects, we do not think that it was contemplated that EPA could alter the conclusion of NAS by. revising the NAS assumptions, or injecting new ones, unless it states its reasons for finding reliability — possibly by challenging the NAS approach in terms of later-acquired research and experience.

These factors combine to convince us that, under our view of Congressional intent, we cannot affirm the EPA’s denial of suspension as stated. That is not necessarily to assume, as at least some petitioners do, that the EPA’s process must be brought to nullity.

The procedures followed in this case, whether or not based on rulings that were “mistaken” when made, have resulted in a record that leaves this court uncertain, at a minimum, whether the essentials of the intention of Congress were achieved. This requires a remand whereby the record as made will be supplemented by further proceedings. In the interest of justice, see 28 U.S.C. § 2106, and mutual regard for Congressional objective, the parties should have opportunity on remand to address themselves to matters not previously put before them by EPA for comment, including material contained in the Technical Appendix filed by EPA in 1972 subsequent to its Decision.

It is contemplated that, in the interest of providing a reasoned decision, the remand proceeding will involve some opportunity for cross-examination. In the remand proceeding — not governed by the same time congestion as the initial Decision process — we require reasonable cross-examination as to new lines of testimony, and as to submissions previously made to EPA in the hearing on a proffer that critical questions could not be satisfactorily pursued by procedures previously in effect. There is, however, still need for expedition, both by virtue of our order and the “lead time” problem, and the EPA may properly confine cross-examination to the essentials, avoiding discursive or repetitive questioning.

Following our suggestion in Environmental Defense Fund, Inc. v. EPA, 150 U.S.App.D.C. 348, 465 F.2d 528 (1972), the Administrator may consider possible use of interim standards short of complete suspension. The statute permits conditioning of suspension on the adoption, by virtue of the information adduced in the suspension proceeding, of *650interim standards, higher than those set for 1974.129

We cannot grant petitioners’ request that this court order a suspension since determinations which Congress made necessary conditions of suspension, as to the public interest and good faith, have not been made by the Administrator. The Administrator’s decision did not reach these questions and accordingly we must remand for further consideration. The initial requirement that an EPA decision on the suspension, aye or nay, be made within 60 days of the application, obviously does not preclude further consideration following remand by the court. In the interest of justice, 28 U.S.C. § 2106, and the Congressional intention that decisions be made timely in the light of considerations of “lead time” for 1975 model year production, we require the suspension deliberations by EPA to be completed within 60 days. The Administrator’s decision on remand must, of course, be consistent with our legal rulings herein— including the need for redefinition of light duty vehicles, and promulgation of an appropriate regulation.

Since our. disposition on remand requires a public interest determination, it disposes of the claim of petitioner Chrysler that the National Environmental Policy Act, 42 U.S.C. § 4321 et seq., requires that an impact statement be filed by the Administrator pursuant to a suspension decision.

The purpose of NEPA is to assure presentation to Congress and the public of the environmental impact of executive action. Here Congress has already decided that the environmental dangers require the statutory standards. The only executive decision is of a one year deferral, and the very stuff of such a decision, at least with a public interest determination, is to assess, inter alia, the environmental consequences of action and inaction. NEPA’s objective will be fully served. As we stated in National Resources Defense Council, Inc. v. Morton, 148 U.S.App.D.C. 5, 15, 458 F.2d 827, 837 (1972), the requirements of NEPA should be subject to a “construction of reasonableness.” Although we do not reach the question whether EPA is automatically and completely exempt from NEPA, we see little need in requiring a NEPA statement from an agency whose raison d’etre is the protection of the environment and whose decision on suspension is necessarily infused with the environmental considerations so pertinent to Congress in designing the statutory framework. To require a “statement,” in addition to a decision setting forth the same considerations, would be a legalism carried to the extreme.

In conformance to the Congressional contemplation of expedition, and our responsibilities as an appellate court, we further require that the Administrator render a decision, on the basis of the best information available, which extends to all the determinations which the statute requires as a condition of suspension.130 We do not preclude further consideration of the question of “available technology,” especially if developments in the art provide enlightenment. Last but not least, especially in view of Ford’s submission and the NAS Report concerning interim standards, we reiterate that the EPA’s determination may consist of a conditional suspension that results in higher standards than an outright grant of applications for suspension.

The case is remanded for further proceedings not inconsistent with this opinion.

BAZELON, Chief Judge

(concurring in result):

Socrates said that wisdom is the recognition of how much one does not know.1 I may be wise if that is wisdom, because *651I recognize that I do not know enough about dynamometer ■ extrapolations,. deterioration factor adjustments, and the like to decide whether or not the government’s approach to these matters was statistically valid. Therein lies my disagreement with the majority.

The court’s opinion today centers on a substantive evaluation of the Administrator’s assumptions and methodology. I do not have the technical know-how to agree or disagree with that evaluation— at least on the basis of the present record. My grounds for remanding the case rest upon the Administrator’s failure to employ a reasonable decision-making process for so critical and complex a matter. At this time I cannot say to what extent I could undertake an evaluation of the Administrator’s findings if they were based on an adequate decisional process.

I cannot believe that Congress intended this court to delve into the substance of the mechanical, statistical, and technological disputes in this case. Senator Cooper, the author of the judicial review provision, stated repeatedly that this court’s role would be to “determine the question of due process.”2 Thus the court’s proper' role is to see to it that the agency provides “a framework for principled decision-making.”2 3 Such a framework necessarily includes the right of interested parties to- confront the agency’s decision and the requirement that the agency set forth with clarity the grounds for its rejection of opposing views.

The majority’s interpretation of the present statute and the administrative precedents would give us no right to establish these procedural guidelines. Their opinion maintains that the strict deadlines in the Clean Air Act preclude any right to challenge the Administrator until after the decision has been made. It indicates that, since this hearing was “rule-making” rather than “adjudicatory”, cross-examination and confrontation are not required under traditional. rules of administrative law.

I understand this viewpoint, but I do not share it. I do not think the authors of the Clean Air Act intended to put such strict limits on our review of the Administrator’s decision-making process. Further, the interests at stake in this case are too important to be resolved on the basis of traditional administrative labels. We recognized two years ago that environmental litigation represents a “new era” in administrative law.4 We are dealing here not with an airline’s fares or a broadcaster’s wattage, but with all humanity’s interest in life, health, and a harmonious relationship with the elements of nature.

This “new era” does not mean that courts will dig deeper into the technical intricacies of an agency’s decision. It means instead that courts will go further in requiring the agency to establish a decision-making process adequate to protect the interests of all “consumers” of the natural environment.5 In some situations, traditional rules of “fairness”— designed only to guard the interests of the specific parties to an agency proceeding — will be inadequate to protect these broader interests. This is such a case. Whether or not traditional admin-' istrative rules require it, the critical character of this decision requires at the *652least a carefully limited right of cross-examination at the hearing and an opportunity to challenge the assumptions and methodology underlying the decision.

The majority’s approach permits the parties to challenge the Administrator’s methodology only through the vehicle of judicial review. I do not think this is an adequate substitute for confrontation prior to the decision. I reach this position not only out of concern for fairness to the parties (“ . . . for if a party first learns of noticed facts through the final report . . . the burden of upsetting a decision announced as final is a heavy one.”8) but also out of awareness of the limits of our own competence for the task. The petitioners’ challenges to the decision force the court to deal with technical intricacies that are beyond our ken.6 7 These complex questions should be resolved in the crucible of debate through the clash of informed but opposing scientific and technological viewpoints.

It is true that courts occasionally find themselves in the thick of technological controversies — e. g., in patent cases. But those are different circumstances. We do not review patent disputes until they have been through a full panoply of procedures involving full rights of confrontation. Further, unlike our decision in a patent case, our decision on the Administrator’s action here is sure to be tested by analysis and challenge in Congress, in the scientific community, and among the public.

My brethren and I are reaching for the same end — a “reasoned decision”'— through different means. They would have us examine the substance of the decision before us. There are some areas of administrative law — involving issues of liberty and individual rights — where judges are on firm ground in undertaking a substantive review of agency action. But in cases of great technological complexity, the best way for courts to guard against unreasonable or erroneous administrative decisions is not for the judges themselves to scrutinize the technical merits of each decision. Rather, it is to establish a decision-making process which assures a reasoned decision that can be held up to the scrutiny of the scientific community and the public.8 “[T]he best test of truth is the power of the thought to get itself accepted in the competition of the market.” 9 If we were to require procedures in this case that open the Administrator’s decision to challenge and force him to respond, we could rely on an informed “market” rather than on our own groping in the dark to test the validity of that decision.

Candor requires the admission that the process of confrontation and challenge might not be possible within the statutory decision period of 60 days. My response would be to permit an extension of the time limit — perhaps 30 days more. This would put less strain on the overall statutory scheme — and on the manufacturers’ lead time — than the months that have been expended in litigation, and now a remand, over the decision. Congress did not intend for us to enforce this relatively minor time restriction so strictly as to do major damage to the statute as a whole.

My brethren argue that the 60-day time limit in the statute precluded any opportunity for cross-examination or confrontation at the time of the original decision. But their opinion would apparently permit these procedural rights on the remand. This bit of judicial legerdemain confounds me. I can find nothing in the statute or common sense *653to support this distinction. If anything, the statute, with its obvious emphasis on reaching a final decision quickly, would dictate procedures at the original decision which were sufficient to produce a reasoned decision without the need for a remand.

Outside of the foregoing differences, I agree with much of the majority opinion. I would have preferred to make the “public interest” factor- — -the considerations set forth in Part III of that opinion — an independent ground for suspension. The court today deals with the public interest indirectly, through the device of burden of proof. I do not fully understand this approach, but I suspect it leads to essentially the same result I favor.

Justice Scalia

delivered the opinion of the Court.

Respondent South Coast Air Quality Management District (District) is a political subdivision of California responsible for air pollution control in the Los Angeles metropolitan area. and parts of surrounding counties that make up the South Coast Air Basin. It enacted six Fleet Rules that generally prohibit the purchase or lease by various public and private *249fleet operators of vehicles that do not comply with stringent emission requirements. The question in this case is whether these local Fleet Rules escape pre-emption under § 209(a) of the Clean Air Act (CAA), 81 Stat. 502, as renumbered and amended, 42 U. S. C. § 7543(a), because they address the purchase of vehicles, rather than their manufacture or sale.

I

The District is responsible under state law for developing and implementing a “comprehensive basinwide air quality management plan” to reduce emission levels and thereby achieve and maintain “state and federal ambient air quality standards.” Cal. Health & Safety Code Ann. § 40402(e) (West 1996). Between June and October 2000, the District adopted six Fleet Rules. The Rules govern operators of fleets of street sweepers (Rule 1186.1), of passenger cars, light-duty trucks, and medium-duty vehicles (Rule 1191), of public transit vehicles and urban buses (Rule 1192), of solid waste collection vehicles (Rule 1193), of airport passenger transportation vehicles, including shuttles and taxicabs picking up airline passengers (Rule 1194), and of heavy-duty on-road vehicles (Rule 1196). All six Rules apply to public operators; three apply to private operators as well (Rules 1186.1, 1193, and 1194).

The Fleet Rules contain detailed prescriptions regarding the types of vehicles that fleet operators must purchase or lease when adding or replacing fleet vehicles. Four of the Rules (1186.1, 1192, 1193, and 1196) require the purchase or lease of “alternative-fuel vehicles,”1 and the other two *250(1191 and 1194) require the purchase or lease of either “alternative-fueled vehicles”2 or vehicles that meet certain emission specifications established by the California Air Resources Board (CARB).3 CARB is a statewide regulatory body that California law designates as “the air pollution control agency for all purposes set forth in federal law.” Cal. *251Health & Safety Code Ann. § 39602 (West 1996). The Rules require operators to keep records of their purchases and leases and provide access to them upon request. See, e. g., Rule 1186.1(g)(1), App. 23. Violations expose fleet operators to fines and other sanctions. See Cal. Health & Safety Code Ann. §§42400-42410, 40447.5 (West 1996 and Supp. 2004).

In August 2000, petitioner Engine Manufacturers Association sued the District and its officials, also respondents, claiming that the Fleet Rules are pre-empted by §209 of the CAA, which prohibits the adoption or attempted enforcement of any state or local “standard relating to the control of emissions from new motor vehicles or new motor vehicle engines.” 42 U. S. C. § 7543(a).4 The District Court granted summary judgment to respondents, upholding the Rules in their entirety. It held that the Rules were not “standard[s]” under § 209(a) because they regulate only the purchase of vehicles that are otherwise certified for sale in California. The District Court recognized that the Courts of Appeals for the First and Second Circuits had previously held that CAA § 209(a) pre-empted state laws mandating that a specified percentage of a manufacturer’s in-state sales be of “zero-emission vehicles.” See Association of Int’l Automobile Mfrs., Inc. v. Commissioner, Mass. Dept. of Environmental Protection, 208 F. 3d 1, 6-7 (CA1 2000); American Automobile Mfrs. Assn. v. Cahill, 152 F. 3d 196, 200 (CA2 1998).5 It did not express disagreement with these rulings, but distinguished them as involving a restriction on vehicle sales rather than vehicle purchases: “Where a state *252regulation does not compel manufacturers to meet a new emissions limit, but rather affects the purchase of vehicles, as the Fleet Rules do, that regulation is not a standard.” 158 F. Supp. 2d 1107, 1118 (CD Cal. 2001).

The Ninth Circuit affirmed on the reasoning of the District Court. 309 F. 3d 550 (2002). We granted certiorari. 539 U. S. 914 (2003).

II

Section 209(a) of the CAA states:

“No State or any political subdivision thereof shall adopt or attempt to enforce any standard relating to the control of emissions from new motor vehicles or new motor vehicle engines subject to this part. No State shall require certification, inspection, or any other approval relating to the control of emissions ... as condition precedent to the initial retail sale, titling (if any), or registration of such motor vehicle, motor vehicle engine, or equipment.” 42 U. S. C. § 7543(a).

The District Court’s determination that this express preemption provision did not invalidate the Fleet Rules hinged on its interpretation of the word “standard” to include only regulations that compel manufacturers to meet specified emission limits. This interpretation of “standard” in turn caused the court to draw a distinction between purchase restrictions (not pre-empted) and sale restrictions (preempted). Neither the manufacturer-specific interpretation of “standard” nor the resulting distinction between purchase and sale restrictions finds support in the text of § 209(a) or the structure of the CAA.

“Statutory construction must begin with the language employed by Congress and the assumption that the ordinary meaning of that language accurately expresses the legislative purpose.” Park ’N Fly, Inc. v. Dollar Park & Fly, Inc., 469 U. S. 189, 194 (1985). Today, as in 1967 when § 209(a) became law, “standard” is defined as that which “is estab*253lished by authority, custom, or general consent, as a model or example; criterion; test.” Webster’s Second New International Dictionary 2455 (1945). The criteria referred to in § 209(a) relate to the emission characteristics of a vehicle or engine. To meet them the vehicle or engine must not emit more than a certain amount of a given pollutant, must be equipped with a certain type of pollution-control device, or must have some other design feature related to the control of emissions. This interpretation is consistent with the use of “standard” throughout Title II of the CAA (which governs emissions from moving sources) to denote requirements such as numerical emission levels with which vehicles or engines must comply, e. g., 42 U. S. C. § 7521(a)(3)(B)(ii), or emission-control technology with which they must be equipped, e. g., § 7521(a)(6).

Respondents, like the courts below, engraft onto this meaning of “standard” a limiting component, defining it as only “[a] production mandat[e] that require[s] manufacturers to ensure that the vehicles they produce have particular emissions characteristics, whether individually or in the aggregate.” Brief for Respondent South Coast Air Quality Management District 13 (emphases added). This confuses standards with the means of enforcing standards. Manufacturers (or purchasers) can be made responsible for ensuring that vehicles comply with emission standards, but the standards themselves are separate from those enforcement techniques. While standards target vehicles or engines, standard-enforcement efforts that are proscribed by §209 can be directed to manufacturers or purchasers.

The distinction between “standards,” on the one hand, and methods of standard enforcement, on the other, is borne out in the provisions immediately following § 202. These separate provisions enforce the emission criteria — i. e., the §202 standards. Section 203 prohibits manufacturers from selling any new motor vehicle that is not covered by a “certificate of conformity.” 42 U. S. C. § 7522(a). Section 206 *254enables manufacturers to obtain such a certificate by demonstrating to the Environmental Protection Agency that their vehicles or engines conform to the §202 standards. §7525. Sections 204 and 205 subject manufacturers, dealers, and others who violate the CAA to fines imposed in civil or administrative enforcement actions. §§7523-7524. By defining “standard” as a “production mandate directed toward manufacturers,” respondents lump together §202 and these other distinct statutory provisions, acknowledging a standard to be such only when it is combined with a mandate that prevents manufacturers from selling noncomplying vehicles.

That a standard is a standard even when not enforced through manufacturer-directed regulation can be seen in Congress’s use of the term in another portion of the CAA. As the District Court recognized, CAA § 246 (in conjunction with its accompanying provisions) requires state-adopted and federally approved “restrictions on the purchase of fleet vehicles to meet clean-air standards” 158 F. Supp. 2d, at 1118 (emphasis added); see also 42 U. S. C. §§7581-7590. (Respondents do not defend the District’s Fleet Rules as authorized by this provision; the Rules do not comply with all of the requirements that it contains.) Clearly, Congress contemplated the enforcement of emission standards through purchase requirements.6

Respondents contend that their qualified meaning of “standard” is necessary to prevent § 209(a) from pre-empting “far too much” by “encompass[ing] a broad range of state-level clean-air initiatives” such as voluntary incentive pro*255grams. Brief for Respondent South Coast Air Quality Management District 29; id., at 29-30. But it is hard to see why limitation to mandates on manufacturers is necessary for this purpose; limitation to mandates on manufacturers and purchasers, or to mandates on anyone, would have the same salvific effect. We need not resolve application of § 209(a) to voluntary incentive programs in this case, since all the Fleet Rules are mandates.

In addition to having no basis in the text of the statute, treating sales restrictions and purchase restrictions differently for pre-emption purposes would make no sense. The manufacturer’s right to sell federally approved vehicles is meaningless in the absence of a purchaser’s right to buy them. It is true that the Fleet Rules at issue here cover only certain purchasers and certain federally certified vehicles, and thus do not eliminate all demand for covered vehicles. But if one State or political subdivision may enact such rules, then so may any other; and the end result would undo Congress’s carefully calibrated regulatory scheme.

A command, accompanied by sanctions, that certain purchasers may buy only vehicles with particular emission characteristics is as much an “attempt to enforce” a “standard” as a command, accompanied by sanctions, that a certain percentage of a manufacturer’s sales volume must consist of such vehicles. We decline to read into § 209(a) a purchase/ sale distinction that is not to be found in the text of § 209(a) or the structure of the CAA.

Ill

The dissent expresses many areas of disagreement with our interpretation, but this should not obscure its agreement with our answer to the question “whether these local Fleet Rules escape pre-emption ... because they address the purchase of vehicles, rather than their manufacture or sale.” Supra, at 249. The dissent joins us in answering “no.” See post, at 262-263 (opinion of Souter, J.). It reaches a differ*256ent outcome in the case because (1) it feels free to read into the unconditional words of the statute a requirement for the courts to determine which purchase restrictions in fact coerce manufacture and which do not; and (2) because it believes that Fleet Rules containing a “commercial availability” proviso do not coerce manufacture.

As to the first point: The language of § 209(a) is categorical. It is (as we have discussed) impossible to find in it an exception for standards imposed through purchase restrictions rather than directly upon manufacturers; it is even more inventive to discover an exception for only that sw&category of standards-imposed-through-purchase-restrictions that does hot coerce manufacture. But even if one accepts that invention, one cannot conclude that these “provisos” save the day. For if a vehicle of the mandated type were commercially available, thus eliminating application of the proviso, the need to sell vehicles to persons governed by the Rule would effectively coerce manufacturers into meeting the artificially created demand. To say, as the dissent does, that this would be merely the consequence of “market demand and free competition,” post, at 263, is fanciful. The demand is a demand, not generated by the market but compelled by the Rules, which in turn effectively compels production. To think that the Rules are invalid until such time as one manufacturer makes a compliant vehicle available, whereupon they become binding, seems to us quite bizarre.

The dissent objects to our interpretive method, which neither invokes the “presumption against preemption” to determine the scope of pre-emption nor delves into legislative history. Post, at 260-261. Application of those methods, on which not all Members of this Court agree, demonstrably makes no difference to resolution of the principal question, which the dissent (after applying them) answers the same as we. As for the additional question that the dissent reaches, we think the same is true: The textual obstacles to the strained interpretation that would validate the Rules by rea*257son of the “commercial availability” provisos are insurmountable — principally, the categorical words of § 209(a). The dissent contends that giving these words their natural meaning of barring implementation of standards at the purchase and sale stage renders superfluous the second sentence of § 209(a), which provides: “No State shall require certification, inspection, or any other approval relating to the control of emissions from any new motor vehicle or new motor vehicle engine as condition precedent to the initial retail sale, titling (if any), or registration of such motor vehicle, motor vehicle engine, or equipment.” 42 U. S. C. § 7543(a). We think it not superfluous, since it makes clear that the term “attempt to enforce” in the first sentence is not limited to the actual imposition of penalties for violation, but includes steps preliminary to that action. Ibid. The sentence is, however, fatal to the dissent’s interpretation of the statute. It categorically prohibits “certification, inspection, or any other approval” as conditions precedent to sale. Why in the world would it do that if it had no categorical objection to standards imposed at the sale stage? Why disable the States from assuring compliance with requirements that they are authorized to impose?

The dissent next charges that our interpretation attributes carelessness to Congress because §246 mandates fleet purchasing restrictions, but does so without specifying “notwithstanding” § 209(a). Post, at 264. That addition might have been nice, but hardly seems necessary. It is obvious, after all, that the principal sales restrictions against which § 209(a) is directed are those requiring compliance with state-imposed standards. What §246 mandates are fleet purchase restrictions under federal standards designed precisely for federally required clean-fuel fleet vehicle programs — which programs, in turn, must be federally approved as meeting detailed federal specifications. It is not surprising that a “notwithstanding” § 209(a) did not come to mind. Far from easting doubt upon our interpretation, § 246 *258is impossible to reconcile with the dissent’s interpretation. The fleet purchase standards it mandates must comply strictly with federal specifications, being neither more lenient nor more demanding. But what is the use of imposing such a limitation if the States are entirely free to impose their own fleet purchase standards with entirely different specifications?

Finally, the dissent says that we should “admit” that our opinion pre-empts voluntary incentive programs. Post, at 265-266. Voluntary programs are not at issue in this case, and are significantly different from eommand-and-control regulation. Suffice it to say that nothing in the present opinion necessarily entails pre-emption of voluntary programs. It is at least arguable that the phrase “adopt or attempt to enforce any standard” refers only to standards that are enforceable — a possibility reinforced by the fact that the prohibition is imposed only on entities (States and political subdivisions) that have power to enforce.

IV

The courts below held all six of the Fleet Rules to be entirely outside the pre-emptive reach of § 209(a) based on reasoning that does not withstand scrutiny. In light of the principles articulated above, it appears likely that at least certain aspects of the Fleet Rules are pre-empted. For example, the District may have attempted to enforce CARB’s ULEV, SULEV, and ZEV standards when, in Rule 1194, it required 50% of new passenger-car and medium-duty-vehicle purchases by private airport-shuttle van operators to “meet ULEV, SULEV, or ZEV emission standards” after July 1, 2001, and 100% to meet those standards after July 1, 2002.7 See Rules 1194(d)(2)(A)-(B), App. 62.

It does not necessarily follow, however, that the Fleet Rules are pre-empted in toto. -We have not addressed a *259number of issues that may affect the ultimate disposition of petitioners’ suit, including the scope of petitioners’ challenge, whether some of the Fleet Rules (or some applications of them) can be characterized as internal state purchase decisions (and, if so, whether a different standard for preemption applies), and whether § 209(a) pre-empts the Fleet Rules even as applied beyond the purchase of new vehicles (e. g., to lease arrangements or to the purchase of used vehicles). These questions were neither passed on below nor presented in the petition for certiorari. They are best addressed in the first instance by the lower courts in light of the principles articulated above.

The judgment is vacated, and the case is remanded for further proceedings consistent with this opinion.

It is so ordered.

Justice Souter,

dissenting.

The Court holds that preemption by the Clean Air Act, 77 Stat. 392, as amended, 42 U. S. C. § 7401 et seq., prohibits one of the most polluted regions in the United States1 from requiring private fleet operators to buy clean engines that are readily available on the commercial market. I respectfully dissent and would hold that the South Coast Air Quality Management District Fleet Rules are not preempted by the Act.

I

So far as it concerns this case, § 209(a) of the Act provides that “[n]o State or any political subdivision thereof shall adopt or attempt to enforce any standard relating to the control of emissions from new motor vehicles or new motor vehicle engines subject to [Title II of the Act].” 42 U. S. C. *260§ 7543(a). The better reading of this provision rests on two interpretive principles the majority opinion does not address.

First, “[i]n all pre-emption cases, and particularly in those [where] Congress has legislated ... in a field which the States have traditionally occupied, we start with the assumption that the historic police powers of the States were not to be superseded by the Federal Act unless that was the clear and manifest purpose of Congress.” Medtronic, Inc. v. Lohr, 518 U. S. 470, 485 (1996) (citation and internal quotation marks omitted); see also Wisconsin Public Intervenor v. Mortier, 501 U. S. 597, 605 (1991) (applying presumption against preemption to a local regulation). The pertinence of this presumption against federal preemption is clear enough from the terms of the Act itself: § 101 states that “air pollution prevention (that is, the reduction or elimination, through any measures, of the amount of pollutants produced or created at the source) and air pollution control at its source is the primary responsibility of States and local governments.” 42 U. S. C. § 7401(a)(3);2 see Huron Portland Cement Co. v. Detroit, 362 U. S. 440, 442 (1960) (“Legislation designed to free from pollution the very air that people breathe clearly falls within the exercise of even the most traditional concept of what is compendiously known as the police power”). The resulting presumption against displacing law enacted or authorized by a State applies both to the “question whether *261Congress intended any pre-emption at all” and to “questions concerning the scope of [§209(a)’s] intended invalidation of state law.” Medtronic, supra, at 485 (emphasis in original).

Second, legislative history should inform interpretive choice, and the legislative history of this preemption provision shows that Congress’s purpose in passing it was to stop States from imposing regulatory requirements that directly limited what manufacturers could sell. During the hearings leading up to the 1967 amendments, “[t]he auto industry .. . was adamant that the nature of their manufacturing mechanism required a single national standard in order to eliminate undue economic strain on the industry.” S. Rep. No. 403, 90th Cong., 1st Sess., 33 (1967). Auto manufacturers sought to safeguard “[t]he ability of those engaged in the manufacture of automobiles to obtain clear and consistent answers concerning emission controls,” and to prevent “a chaotic situation from developing in interstate commerce in new motor vehicles.” H. R. Rep. No. 728, 90th Cong., 1st Sess., 21 (1967). Cf. Air Pollution Control, Hearings on S. 306 before a Special Subcommittee on Air and Water Pol-, lution of the Senate Committee on Public Works, 89th Cong., 1st Sess., 91 (1965) (Sen. Muskie) (“Do you think a given manufacturer could produce automobiles meeting 50 standards?”). Congress was not responding to concerns about varying regional appetites for whatever vehicle models the manufacturers did produce; it was addressing the industry’s fear that States would bar manufacturers from selling engines that failed to meet specifications that might be different in each State.3

*262Section 209(a) can easily be read to give full effect to both principles. As amended in 1967, § 202 of the Act authorized federal regulators to promulgate emissions standards for “any class or classes of new motor vehicles or new motor vehicle engines.” § 202(a), 81 Stat. 499. The 1967 amendments in turn defined “new motor vehicle” as “a motor vehicle the equitable or legal title to which has never been transferred to an ultimate purchaser,” and a “new motor vehicle engine” as “an engine in a new motor vehicle or a motor vehicle engine the equitable or legal title to which has never been transferred to the ultimate purchaser.” §212(3), 81 Stat. 503. Section 202 of the 1967 Act, in other words, is naturally understood as concerning itself with vehicles prior to sale and eligible to be sold. Section 203 further underscored this focus on what manufacturers could produce for sale: as incorporated in the 1967 amendments, §203 prohibited a variety of acts by manufacturers, but left vehicle purchasers and users entirely unregulated. 81 Stat. 499.

On this permissible reading of the 1967 amendments, § 209(a) has no preemptive application to South Coast’s fleet purchase requirement. The National Government took over the direct regulation of manufacturers’ design specifications addressing tailpipe emissions, and disabled States (the California exception aside, see n. 3, supra) from engaging in the same project. The “standards” that § 209(a) preempts, accordingly, are production mandates imposed directly on manufacturers as a condition of sale. Section 209(a) simply does not speak to regulations that govern a vehicle buyer’s choice between various commercially available options.

This is not to say that every conceiváble purchase restriction would be categorically free from preemption. A state law prohibiting any purchase by any buyer of any vehicle that failed to meet novel, state-specified emissions criteria would have the same effect as direct regulation of car manufacturers, and would be preempted by § 209(a) as an “attempt to enforce [a] standard relating to the control of emissions *263from new motor vehicles.” 42 U. S. C. § 7543(a). But that fantasy is of no concern here, owing to a third central point that the majority passes over: South Coast’s Fleet Rules require the purchase of cleaner engines only if cleaner engines are commercially available. E.g., App. 69 (Fleet Rule 1196(e)(1)(C) (exempting fleets from Rule if no complying engine “is commercially available from any manufacturer . . . or could be used in a specific application”)); see also App. 21, 30, 50, 55, 63 (Fleet Rules 1186.1(e), 1191(f)(8), 1192(e)(2), 1193(e)(3), and 1194(e)(2)). If no one is selling cleaner engines, fleet owners are free to buy any vehicles they desire. The manufacturers would, of course, understand that a market existed for cleaner engines, and if one auto maker began producing them, others might well be induced to do the same; but that would not matter under the Act, which was not adopted to exempt producers from market demand and free competition. So long as a purchase requirement is subject to a commercial availability proviso, there is no basis to condemn that kind of market-based limitation along with the state command-and-control regulation of production specifications that prompted the passage of §209.

In sum, I am reading “standard” in a practical way that keeps the Act’s preemption of standards in tune with Congress’s object in providing for preemption, which was to prevent the States from forcing manufacturers to produce engines with particular characteristics as a legal condition of sale. The majority’s approach eliminates this consideration of legislative purposes, as well as the presumption against preemption, by acting as though anything that could possibly be described as a standard must necessarily be a “standard” for the purposes of the Act: a standard is a standard is a standard.4 The majority reveals its misalliance with Ger*264trude Stein throughout its response to this dissent. See ante, at 256-257, 258.

II

Reading the statute this way not only does a better job of honoring preemption principles consistently with congressional intent, but avoids some difficulties on the majority’s contrary interpretation. To begin with, the Court’s broad definition of an “ ‘attempt to enforce any standard relating to the control of emissions,’ ” ante, at 252, renders superfluous the second sentence of § 209(a), which provides that “[n]o State shall require certification, inspection, or any other approval relating to the control of emissions from any new motor vehicle ... as condition precedent to the initial retail sale, titling (if any), or registration of such motor vehicle,” 42 U. S. C. § 7543(a). At the very least, on the majority’s view, it is hard to imagine any state inspection requirement going to the control of emissions from a new motor vehicle that would not be struck down anyway as an attempt to enforce an emissions standard.

Next, on the majority’s broad interpretation of “standard,” Congress would seem to have been careless in drafting a critical section of the Act. In the one clear instance of which we are aware in which the Act authorizes States to enact laws that would otherwise be preempted by § 209, Congress expressly provided that the authorization is effective notwithstanding that preemption section. See 42 U. S. C. §7507 (authorizing States to adopt California production mandates “[Notwithstanding section 7543(a) of this title”). The natural negative implication is that, if a statutory authorization does not include such a “notwithstanding” clause or something similar, its subject matter would not otherwise be preempted by § 209(a). Given that, the majority’s interpretation of the scope of § 209(a) is difficult to square with § 246, which requires States to establish fleet purchasing requirements for “covered fleet operator[s]” in ozone and carbon monoxide “nonattainment areas” (that is, regions strug*265gling with especially intractable pollution), 42 U. S. C. § 7586. Section 246 thus requires States, in some cases, to establish precisely the kind of purchaser regulations (adopted here by a lower level governmental authority) that the majority claims have been preempted by § 209(a). But § 246 gives no indication that its subject matter would otherwise be preempted; there is certainly no “notwithstanding” clause. This silence suggests that Congress never thought § 209(a) would have any preemptive effect on fleet purchasing requirements like the ones at issue.

Finally, the Court suggests that both voluntary incentive programs, ante, at 254-255, and internal state purchasing decisions, ante, at 258-259, may well be permissible on its reading of § 209(a). These suggestions are important in avoiding apparent implausibility in the majority’s position; if a State were said to be barred even from deciding to run a cleaner fleet than the National Government required, it would take an airtight argument to convince anyone that Congress could have meant such a thing. But it is difficult, when actually applying the majority’s expansive sense of forbidden “standard,” to explain how the specification of emissions characteristics in a State’s internal procurement guidelines could escape being considered an impermissible “adoption of a] standard,” 42 U. S. C. § 7543(a), even if the standard only guided local purchasing decisions. By the same token, it is not obvious how, without some legal sleight of hand, the majority can avoid preempting voluntary incentive programs aimed at the private sector; the benefit proffered by such schemes hinges on the recipient’s willingness' to buy a vehicle or engine that complies with an emissions standard (i. e., a vehicle or engine that, in the words of the majority, “must not emit more than a certain amount of a given pollutant, must be equipped with a certain type of pollution-control device, or must have some other design feature related to the control of emissions,” ante, at 253). Such a program clearly “adopt[s]” an emissions standard as *266the majority defines it. Cf. ibid, (cautioning respondents not to “confus[e] standards with the means of enforcing standards”). The Court should, then, admit to preemption of state programs that even petitioners concede are not barred by § 209(a). See Reply Brief for Petitioners 7 (acknowledging that § 209(a) does not preempt voluntary incentive programs). That is not a strong recommendation for the majority’s reading.

Ill

These objections to the Court’s interpretation are not, to be sure, dispositive, standing alone. They call attention to untidy details, and rightly understood legislation can be untidy: statutes can be unsystematic, redundant, and fuzzy about drawing lines. As a purely textual matter, both the majority’s reading and mine have strengths and weaknesses. The point is that the tie breakers cut in favor of sustaining the South Coast Fleet Rules. My reading adheres more closely to the legislative history of § 209(a). It takes proper account of the fact that the Fleet Rules with this commercial availability condition do not require manufacturers, even indirectly, to produce a new kind of engine. And, most importantly, my reading adheres to the well-established presumption against preemption.