Justice SCALIA delivered the opinion of the Court.

The Clean Air Act directs the Environmental Protection Agency to regulate emissions of hazardous air pollutants from power plants if the Agency finds regulation "appropriate and necessary." We must decide whether it was reasonable for EPA to refuse to consider cost when making this finding.

I

The Clean Air Act establishes a series of regulatory programs to control air pollution from stationary sources (such as refineries and factories) and moving sources (such as cars and airplanes). 69 Stat. 322, as amended, 42 U.S.C. §§ 7401-7671q. One of these is the National Emissions Standards for Hazardous Air Pollutants Program — the hazardous-air-pollutants program, for short. Established in its current form by the Clean Air Act Amendments of 1990, 104 Stat. 2531, this program targets for regulation stationary-source emissions of more than 180 specified "hazardous air pollutants." § 7412(b).

For stationary sources in general, the applicability of the program depends in part on how much pollution the source emits. A source that emits more than 10 tons of a single pollutant or more than 25 tons of a combination of pollutants per year is called a major source. § 7412(a)(1). EPA is required to regulate all major sources under the program. § 7412(c)(1)-(2). A source whose emissions do not cross the just-mentioned thresholds is called an area source. § 7412(a)(2). The Agency is required to regulate an area source under the program if it "presents a threat of adverse effects to human health or the environment ... warranting regulation." § 7412(c)(3).

At the same time, Congress established a unique procedure to determine the applicability of the program to fossil-fuel-fired power plants. The Act refers to these plants as electric utility steam generating units, but we will simply call them power plants. Quite apart from the hazardous-air-pollutants program, the Clean Air Act Amendments of 1990 subjected power plants to various regulatory requirements. The parties agree that these requirements were expected to have the collateral effect of reducing power plants' emissions of hazardous air pollutants, although the extent of the reduction was unclear. Congress directed the Agency to "perform a study of the hazards to public health reasonably anticipated to occur as a result of emissions by [power plants] of [hazardous air pollutants] after imposition of the requirements of this chapter." § 7412(n)(1)(A). If the Agency "finds ... regulation is appropriate and necessary after considering the results of the study," it "shall regulate [power plants] under [§ 7412]." Ibid. EPA has interpreted the Act to mean that power plants become subject to regulation on the same terms as ordinary major and area sources, see 77 Fed.Reg. 9330 (2012), and we assume without deciding that it was correct to do so.

And what are those terms? EPA must first divide sources covered by the program into categories and subcategories in accordance with statutory criteria. § 7412(c)(1). For each category or subcategory, the Agency must promulgate certain minimum emission regulations, known as floor standards. § 7412(d)(1), (3). The statute generally calibrates the floor standards to reflect the emissions limitations already achieved by the best-performing 12% of sources within the category or subcategory. § 7412(d)(3). In some circumstances, the Agency may also impose more stringent emission regulations, known as beyond-the-floor standards. The statute expressly requires the Agency to consider cost (alongside other specified factors) when imposing beyond-the-floor standards. § 7412(d)(2).

EPA completed the study required by § 7412(n)(1)(A) in 1998, 65 Fed.Reg. 79826 (2000), and concluded that regulation of coal- and oil-fired power plants was "appropriate and necessary" in 2000, id., at 79830. In 2012, it reaffirmed the appropriate-and-necessary finding, divided power plants into subcategories, and promulgated floor standards. The Agency found regulation "appropriate" because (1) power plants' emissions of mercury and other hazardous air pollutants posed risks to human health and the environment and (2) controls were available to reduce these emissions. 77 Fed.Reg. 9363. It found regulation "necessary" because the imposition of the Act's other requirements did not eliminate these risks. Ibid. EPA concluded that "costs should not be considered" when deciding whether power plants should be regulated under § 7412. Id., at 9326.

In accordance with Executive Order, the Agency issued a "Regulatory Impact Analysis" alongside its regulation. This analysis estimated that the regulation would force power plants to bear costs of $9.6 billion per year. Id., at 9306. The Agency could not fully quantify the benefits of reducing power plants' emissions of hazardous air pollutants; to the extent it could, it estimated that these benefits were worth $4 to $6 million per year. Ibid. The costs to power plants were thus between 1,600 and 2,400 times as great as the quantifiable benefits from reduced emissions of hazardous air pollutants. The Agency continued that its regulations would have ancillary benefits — including cutting power plants' emissions of particulate matter and sulfur dioxide, substances that are not covered by the hazardous-air-pollutants program. Although the Agency's appropriate-and-necessary finding did not rest on these ancillary effects, id., at 9320, the regulatory impact analysis took them into account, increasing the Agency's estimate of the quantifiable benefits of its regulation to $37 to $90 billion per year, id., at 9306. EPA concedes that the regulatory impact analysis "played no role" in its appropriate-and-necessary finding. Brief for Federal Respondents 14.

Petitioners (who include 23 States) sought review of EPA's rule in the Court of Appeals for the D.C. Circuit. As relevant here, they challenged the Agency's refusal to consider cost when deciding whether to regulate power plants. The Court of Appeals upheld the Agency's decision not to consider cost, with Judge Kavanaugh concurring in part and dissenting in part. White Stallion Energy Center, LLC v. EPA, 748 F.3d 1222 (2014) (per curiam). We granted certiorari. 574 U.S. ___, 135 S.Ct. 702, 703, 190 L.Ed.2d 434 (2014).

II

Federal administrative agencies are required to engage in "reasoned decisionmaking." Allentown Mack Sales & Service, Inc. v. NLRB, 522 U.S. 359, 374, 118 S.Ct. 818, 139 L.Ed.2d 797 (1998) (internal quotation marks omitted). "Not only must an agency's decreed result be within the scope of its lawful authority, but the process by which it reaches that result must be logical and rational." Ibid. It follows that agency action is lawful only if it rests "on a consideration of the relevant factors." Motor Vehicle Mfrs. Assn. of United States, Inc. v. State Farm Mut. Automobile Ins. Co., 463 U.S. 29, 43, 103 S.Ct. 2856, 77 L.Ed.2d 443 (1983) (internal quotation marks omitted).

EPA's decision to regulate power plants under § 7412 allowed the Agency to reduce power plants' emissions of hazardous air pollutants and thus to improve public health and the environment. But the decision also ultimately cost power plants, according to the Agency's own estimate, nearly $10 billion a year. EPA refused to consider whether the costs of its decision outweighed the benefits. The Agency gave cost no thought at all, because it considered cost irrelevant to its initial decision to regulate.

EPA's disregard of cost rested on its interpretation of § 7412(n)(1)(A), which, to repeat, directs the Agency to regulate power plants if it "finds such regulation is appropriate and necessary." The Agency accepts that it could have interpreted this provision to mean that cost is relevant to the decision to add power plants to the program. Tr. of Oral Arg. 44. But it chose to read the statute to mean that cost makes no difference to the initial decision to regulate. See 76 Fed.Reg. 24988 (2011) ("We further interpret the term `appropriate' to not allow for the consideration of costs"); 77 Fed.Reg. 9327 ("Cost does not have to be read into the definition of `appropriate'").

We review this interpretation under the standard set out in Chevron U.S.A. Inc. v. Natural Resources Defense Council, Inc., 467 U.S. 837, 104 S.Ct. 2778, 81 L.Ed.2d 694 (1984). Chevron directs courts to accept an agency's reasonable resolution of an ambiguity in a statute that the agency administers. Id., at 842-843, 104 S.Ct. 2778. Even under this deferential standard, however, "agencies must operate within the bounds of reasonable interpretation." Utility Air Regulatory Group v. EPA, 573 U.S. ___, ___, 134 S.Ct. 2427, 2442, 189 L.Ed.2d 372 (2014) (internal quotation marks omitted). EPA strayed far beyond those bounds when it read § 7412(n)(1) to mean that it could ignore cost when deciding whether to regulate power plants.

A

The Clean Air Act treats power plants differently from other sources for purposes of the hazardous-air-pollutants program. Elsewhere in § 7412, Congress established cabined criteria for EPA to apply when deciding whether to include sources in the program. It required the Agency to regulate sources whose emissions exceed specified numerical thresholds (major sources). It also required the Agency to regulate sources whose emissions fall short of these thresholds (area sources) if they "presen[t] a threat of adverse effects to human health or the environment... warranting regulation." § 7412(c)(3). In stark contrast, Congress instructed EPA to add power plants to the program if (but only if) the Agency finds regulation "appropriate and necessary." § 7412(n)(1)(A). One does not need to open up a dictionary in order to realize the capaciousness of this phrase. In particular, "appropriate" is "the classic broad and all-encompassing term that naturally and traditionally includes consideration of all the relevant factors." 748 F.3d, at 1266 (opinion of Kavanaugh, J.). Although this term leaves agencies with flexibility, an agency may not "entirely fai[l] to consider an important aspect of the problem" when deciding whether regulation is appropriate. State Farm, supra, at 43, 103 S.Ct. 2856.

Read naturally in the present context, the phrase "appropriate and necessary" requires at least some attention to cost. One would not say that it is even rational, never mind "appropriate," to impose billions of dollars in economic costs in return for a few dollars in health or environmental benefits. In addition, "cost" includes more than the expense of complying with regulations; any disadvantage could be termed a cost. EPA's interpretation precludes the Agency from considering any type of cost — including, for instance, harms that regulation might do to human health or the environment. The Government concedes that if the Agency were to find that emissions from power plants do damage to human health, but that the technologies needed to eliminate these emissions do even more damage to human health, it would still deem regulation appropriate. See Tr. of Oral Arg. 70. No regulation is "appropriate" if it does significantly more harm than good.

There are undoubtedly settings in which the phrase "appropriate and necessary" does not encompass cost. But this is not one of them. Section 7412(n)(1)(A) directs EPA to determine whether "regulation is appropriate and necessary." (Emphasis added.) Agencies have long treated cost as a centrally relevant factor when deciding whether to regulate. Consideration of cost reflects the understanding that reasonable regulation ordinarily requires paying attention to the advantages and the disadvantages of agency decisions. It also reflects the reality that "too much wasteful expenditure devoted to one problem may well mean considerably fewer resources available to deal effectively with other (perhaps more serious) problems." Entergy Corp. v. Riverkeeper, Inc., 556 U.S. 208, 233, 129 S.Ct. 1498, 173 L.Ed.2d 369 (2009) (BREYER, J., concurring in part and dissenting in part). Against the backdrop of this established administrative practice, it is unreasonable to read an instruction to an administrative agency to determine whether "regulation is appropriate and necessary" as an invitation to ignore cost.

Statutory context reinforces the relevance of cost. The procedures governing power plants that we consider today appear in § 7412(n)(1), which bears the caption "Electric utility steam generating units." In subparagraph (A), the part of the law that has occupied our attention so far, Congress required EPA to study the hazards to public health posed by power plants and to determine whether regulation is appropriate and necessary. But in subparagraphs (B) and (C), Congress called for two additional studies. One of them, a study into mercury emissions from power plants and other sources, must consider "the health and environmental effects of such emissions, technologies which are available to control such emissions, and the costs of such technologies." § 7412(n)(1)(B) (emphasis added). This directive to EPA to study cost is a further indication of the relevance of cost to the decision to regulate.

In an effort to minimize this express reference to cost, EPA now argues that § 7412(n)(1)(A) requires it to consider only the study mandated by that provision, not the separate mercury study, before deciding whether to regulate power plants. But when adopting the regulations before us, the Agency insisted that the provisions concerning all three studies "provide a framework for [EPA's] determination of whether to regulate [power plants]." 76 Fed.Reg. 24987. It therefore decided "to interpret the scope of the appropriate and necessary finding in the context of all three studies." 77 Fed.Reg. 9325 (emphasis added). For example:

EPA has not explained why § 7412(n)(1)(B)'s reference to "environmental effects ... and ... costs" provides "direct evidence that Congress was concerned with environmental effects," but not "direct evidence" that it was concerned with cost. Chevron allows agencies to choose among competing reasonable interpretations of a statute; it does not license interpretive gerrymanders under which an agency keeps parts of statutory context it likes while throwing away parts it does not.

B

EPA identifies a handful of reasons to interpret § 7412(n)(1)(A) to mean that cost is irrelevant to the initial decision to regulate. We find those reasons unpersuasive.

EPA points out that other parts of the Clean Air Act expressly mention cost, while § 7412(n)(1)(A) does not. But this observation shows only that § 7412(n)(1)(A)'s broad reference to appropriateness encompasses multiple relevant factors (which include but are not limited to cost); other provisions' specific references to cost encompass just cost. It is unreasonable to infer that, by expressly making cost relevant to other decisions, the Act implicitly makes cost irrelevant to the appropriateness of regulating power plants. (By way of analogy, the Fourth Amendment's Reasonableness Clause requires searches to be "[r]easonable," while its Warrant Clause requires warrants to be supported by "probable cause." Nobody would argue that, by expressly making level of suspicion relevant to the validity of a warrant, the Fourth Amendment implicitly makes level of suspicion categorically irrelevant to the reasonableness of a search. To the contrary, all would agree that the expansive word "reasonable" encompasses degree of suspicion alongside other relevant circumstances.) Other parts of the Clean Air Act also expressly mention environmental effects, while § 7412(n)(1)(A) does not. Yet that did not stop EPA from deeming environmental effects relevant to the appropriateness of regulating power plants.

Along similar lines, EPA seeks support in this Court's decision in Whitman v. American Trucking Assns., Inc., 531 U.S. 457, 121 S.Ct. 903, 149 L.Ed.2d 1 (2001). There, the Court addressed a provision of the Clean Air Act requiring EPA to set ambient air quality standards at levels "requisite to protect the public health" with an "adequate margin of safety." 42 U.S.C. § 7409(b). Read naturally, that discrete criterion does not encompass cost; it encompasses health and safety. The Court refused to read that provision as carrying with it an implicit authorization to consider cost, in part because authority to consider cost had "elsewhere, and so often, been expressly granted." 531 U.S., at 467, 121 S.Ct. 903. American Trucking thus establishes the modest principle that where the Clean Air Act expressly directs EPA to regulate on the basis of a factor that on its face does not include cost, the Act normally should not be read as implicitly allowing the Agency to consider cost anyway. That principle has no application here. "Appropriate and necessary" is a far more comprehensive criterion than "requisite to protect the public health"; read fairly and in context, as we have explained, the term plainly subsumes consideration of cost.

Turning to the mechanics of the hazardous-air-pollutants program, EPA argues that it need not consider cost when first deciding whether to regulate power plants because it can consider cost later when deciding how much to regulate them. The question before us, however, is the meaning of the "appropriate and necessary" standard that governs the initial decision to regulate. And as we have discussed, context establishes that this expansive standard encompasses cost. Cost may become relevant again at a later stage of the regulatory process, but that possibility does not establish its irrelevance at this stage. In addition, once the Agency decides to regulate power plants, it must promulgate certain minimum or floor standards no matter the cost (here, nearly $10 billion a year); the Agency may consider cost only when imposing regulations beyond these minimum standards. By EPA's logic, someone could decide whether it is "appropriate" to buy a Ferrari without thinking about cost, because he plans to think about cost later when deciding whether to upgrade the sound system.

EPA argues that the Clean Air Act makes cost irrelevant to the initial decision to regulate sources other than power plants. The Agency claims that it is reasonable to interpret § 7412(n)(1)(A) in a way that "harmonizes" the program's treatment of power plants with its treatment of other sources. This line of reasoning overlooks the whole point of having a separate provision about power plants: treating power plants differently from other stationary sources. Congress crafted narrow standards for EPA to apply when deciding whether to regulate other sources; in general, these standards concern the volume of pollution emitted by the source, § 7412(c)(1), and the threat posed by the source "to human health or the environment," § 7412(c)(3). But Congress wrote the provision before us more expansively, directing the Agency to regulate power plants if "appropriate and necessary." "That congressional election settles this case. [The Agency's] preference for symmetry cannot trump an asymmetrical statute." CSX Transp., Inc. v. Alabama Dept. of Revenue, 562 U.S. 277, 296, 131 S.Ct. 1101, 179 L.Ed.2d 37 (2011).

EPA persists that Congress treated power plants differently from other sources because of uncertainty about whether regulation of power plants would still be needed after the application of the rest of the Act's requirements. That is undoubtedly one of the reasons Congress treated power plants differently; hence § 7412(n)(1)(A)'s requirement to study hazards posed by power plants' emissions "after imposition of the requirements of [the rest of the Act]." But if uncertainty about the need for regulation were the only reason to treat power plants differently, Congress would have required the Agency to decide only whether regulation remains "necessary," not whether regulation is "appropriate and necessary." In any event, EPA stated when it adopted the rule that "Congress did not limit [the] appropriate and necessary inquiry to [the study mentioned in § 7412(n)(1)(A)]." 77 Fed.Reg. 9325. The Agency instead decided that the appropriate-and-necessary finding should be understood in light of all three studies required by § 7412(n)(1), and as we have discussed, one of those three studies reflects concern about cost.

C

The dissent does not embrace EPA's far-reaching claim that Congress made costs altogether irrelevant to the decision to regulate power plants. Instead, it maintains that EPA need not "explicitly analyze costs" before deeming regulation appropriate, because other features of the regulatory program will on their own ensure the cost-effectiveness of regulation. Post, at 2714 (opinion of KAGAN, J.). This line of reasoning contradicts the foundational principle of administrative law that a court may uphold agency action only on the grounds that the agency invoked when it took the action. SEC v. Chenery Corp., 318 U.S. 80, 87, 63 S.Ct. 454, 87 S.Ct. 626 (1943). When it deemed regulation of power plants appropriate, EPA said that cost was irrelevant to that determination — not that cost-benefit analysis would be deferred until later. Much less did it say (what the dissent now concludes) that the consideration of cost at subsequent stages will ensure that the costs are not disproportionate to the benefits. What it said is that cost is irrelevant to the decision to regulate.

That is enough to decide these cases. But for what it is worth, the dissent vastly overstates the influence of cost at later stages of the regulatory process. For example, the dissent claims that the floor standards — which the Act calibrates to reflect emissions limitations already achieved by the best-performing sources in the industry — reflect cost considerations, because the best-performing power plants "must have considered costs in arriving at their emissions outputs." Post, at 2719. EPA did not rely on this argument, and it is not obvious that it is correct. Because power plants are regulated under other federal and state laws, the best-performing power plants' emissions limitations might reflect cost-blind regulation rather than cost-conscious decisions. Similarly, the dissent suggests that EPA may consider cost when dividing sources into categories and subcategories. Post, at 2720. Yet according to EPA, "it is not appropriate to premise subcategorization on costs." 77 Fed.Reg. 9395 (emphasis added). That statement presumably explains the dissent's carefully worded observation that EPA considered "technological, geographic, and other factors" when drawing categories, post, at 2720, n. 4, which factors were in turn "related to costs" in some way, post, at 2719. Attenuated connections such as these hardly support the assertion that EPA's regulatory process featured "exhaustive consideration of costs," post, at 2714.

All in all, the dissent has at most shown that some elements of the regulatory scheme mitigate cost in limited ways; it has not shown that these elements ensure cost-effectiveness. If (to take a hypothetical example) regulating power plants would yield $5 million in benefits, the prospect of mitigating cost from $11 billion to $10 billion at later stages of the program would not by itself make regulation appropriate. In all events, we need not pursue these points, because EPA did not say that the parts of the regulatory program mentioned by the dissent prevent the imposition of costs far in excess of benefits. "[EPA's] action must be measured by what [it] did, not by what it might have done." Chenery, supra, at 93-94, 63 S.Ct. 454.

D

Our reasoning so far establishes that it was unreasonable for EPA to read § 7412(n)(1)(A) to mean that cost is irrelevant to the initial decision to regulate power plants. The Agency must consider cost — including, most importantly, cost of compliance — before deciding whether regulation is appropriate and necessary. We need not and do not hold that the law unambiguously required the Agency, when making this preliminary estimate, to conduct a formal cost-benefit analysis in which each advantage and disadvantage is assigned a monetary value. It will be up to the Agency to decide (as always, within the limits of reasonable interpretation) how to account for cost.

Some of the respondents supporting EPA ask us to uphold EPA's action because the accompanying regulatory impact analysis shows that, once the rule's ancillary benefits are considered, benefits plainly outweigh costs. The dissent similarly relies on these ancillary benefits when insisting that "the outcome here [was] a rule whose benefits exceed its costs." Post, at 2722. As we have just explained, however, we may uphold agency action only upon the grounds on which the agency acted. Even if the Agency could have considered ancillary benefits when deciding whether regulation is appropriate and necessary — a point we need not address — it plainly did not do so here. In the Agency's own words, the administrative record "utterly refutes [the] assertion that [ancillary benefits] form the basis for the appropriate and necessary finding." 77 Fed.Reg. 9323. The Government concedes, moreover, that "EPA did not rely on the [regulatory impact analysis] when deciding to regulate power plants," and that "[e]ven if EPA had considered costs, it would not necessarily have adopted ... the approach set forth in [that analysis]." Brief for Federal Respondents 53-54.

We hold that EPA interpreted § 7412(n)(1)(A) unreasonably when it deemed cost irrelevant to the decision to regulate power plants. We reverse the judgment of the Court of Appeals for the D.C. Circuit and remand the cases for further proceedings consistent with this opinion.

It is so ordered.

Justice THOMAS, concurring.

The Environmental Protection Agency (EPA) asks the Court to defer to its interpretation of the phrase "appropriate and necessary" in § 112(n)(1)(A) of the Clean Air Act, 42 U.S.C. § 7412. Justice SCALIA's opinion for the Court demonstrates why EPA's interpretation deserves no deference under our precedents. I write separately to note that its request for deference raises serious questions about the constitutionality of our broader practice of deferring to agency interpretations of federal statutes. See Chevron U.S.A. Inc. v. Natural Resources Defense Council, Inc., 467 U.S. 837, 104 S.Ct. 2778, 81 L.Ed.2d 694 (1984).

Chevron deference is premised on "a presumption that Congress, when it left ambiguity in a statute meant for implementation by an agency, understood that the ambiguity would be resolved, first and foremost, by the agency, and desired the agency (rather than the courts) to possess whatever degree of discretion the ambiguity allows." Smiley v. Citibank (South Dakota), N. A., 517 U.S. 735, 740-741, 116 S.Ct. 1730, 135 L.Ed.2d 25 (1996). We most often describe Congress' supposed choice to leave matters to agency discretion as an allocation of interpretive authority. See, e.g., National Cable & Telecommunications Assn. v. Brand X Internet Services, 545 U.S. 967, 983, 125 S.Ct. 2688, 162 L.Ed.2d 820 (2005) (referring to the agency as "the authoritative interpreter (within the limits of reason) of [ambiguous] statutes"). But we sometimes treat that discretion as though it were a form of legislative power. See, e.g., United States v. Mead Corp., 533 U.S. 218, 229, 121 S.Ct. 2164, 150 L.Ed.2d 292 (2001) (noting that the agency "speak[s] with the force of law when it addresses ambiguity in the statute or fills a space in the enacted law" even when "`Congress did not actually have an intent' as to a particular result"). Either way, Chevron deference raises serious separation-of-powers questions.

As I have explained elsewhere, "[T]he judicial power, as originally understood, requires a court to exercise its independent judgment in interpreting and expounding upon the laws." Perez v. Mortgage Bankers Assn., 575 U.S. ___, ___ 135 S.Ct. 1199, 1217, 191 L.Ed.2d 186 (2015) (opinion concurring in judgment). Interpreting federal statutes — including ambiguous ones administered by an agency — "calls for that exercise of independent judgment." Id., at ___, 135 S.Ct., at 1219. Chevron deference precludes judges from exercising that judgment, forcing them to abandon what they believe is "the best reading of an ambiguous statute" in favor of an agency's construction. Brand X, supra, at 983, 125 S.Ct. 2688. It thus wrests from Courts the ultimate interpretative authority to "say what the law is," Marbury v. Madison, 1 Cranch 137, 177, 2 S.Ct. 60 (1803), and hands it over to the Executive. See Brand X, supra, at 983, 125 S.Ct. 2688 (noting that the judicial construction of an ambiguous statute is "not authoritative"). Such a transfer is in tension with Article III's Vesting Clause, which vests the judicial power exclusively in Article III courts, not administrative agencies. U.S. Const., Art. III, § 1.

In reality, as the Court illustrates in the course of dismantling EPA's interpretation of § 112(n)(1)(A), agencies "interpreting" ambiguous statutes typically are not engaged in acts of interpretation at all. See, e.g., ante, at 2708. Instead, as Chevron itself acknowledged, they are engaged in the "`formulation of policy.'" 467 U.S., at 843, 104 S.Ct. 2778. Statutory ambiguity thus becomes an implicit delegation of rule-making authority, and that authority is used not to find the best meaning of the text, but to formulate legally binding rules to fill in gaps based on policy judgments made by the agency rather than Congress.

Although acknowledging this fact might allow us to escape the jaws of Article III's Vesting Clause, it runs headlong into the teeth of Article I's, which vests "[a]ll legislative Powers herein granted" in Congress. U.S. Const., Art. I, § 1. For if we give the "force of law" to agency pronouncements on matters of private conduct as to which "`Congress did not actually have an intent,'" Mead, supra, at 229, 121 S.Ct. 2164, we permit a body other than Congress to perform a function that requires an exercise of the legislative power. See Department of Transportation v. Association of American Railroads, 575 U.S. ___, ___ _ ___, 135 S.Ct. 1225, 1251-1252, 191 L.Ed.2d 153 (2015) (THOMAS, J., concurring in judgment).

These cases bring into bold relief the scope of the potentially unconstitutional delegations we have come to countenance in the name of Chevron deference. What EPA claims for itself here is not the power to make political judgments in implementing Congress' policies, nor even the power to make tradeoffs between competing policy goals set by Congress, American Railroads, supra, at ___ _ ___, 135 S.Ct., at 1250-1252 (opinion of THOMAS, J.) (collecting cases involving statutes that delegated this legislative authority). It is the power to decide — without any particular fidelity to the text — which policy goals EPA wishes to pursue. Should EPA wield its vast powers over electric utilities to protect public health? A pristine environment? Economic security? We are told that the breadth of the word "appropriate" authorizes EPA to decide for itself how to answer that question. Compare 77 Fed. Reg. 9327 (2012) ("[N]othing about the definition [of "appropriate"] compels a consideration of costs" (emphasis added)) with Tr. of Oral Arg. 42 ("[T]he phrase appropriate and necessary doesn't, by its terms, preclude the EPA from considering cost" (emphasis added)).1

Perhaps there is some unique historical justification for deferring to federal agencies, see Mead, supra, at 243, 121 S.Ct. 2164 (SCALIA, J., dissenting), but these cases reveal how paltry an effort we have made to understand it or to confine ourselves to its boundaries. Although we hold today that EPA exceeded even the extremely permissive limits on agency power set by our precedents, we should be alarmed that it felt sufficiently emboldened by those precedents to make the bid for deference that it did here.2 As in other areas of our jurisprudence concerning administrative agencies, see, e.g., B & B Hardware, Inc. v. Hargis Industries, Inc., 575 U.S. ___, ___ _ ___, 135 S.Ct. 1293, 1315 ___ 1318, 191 L.Ed.2d 222 (2015) (THOMAS, J., dissenting), we seem to be straying further and further from the Constitution without so much as pausing to ask why. We should stop to consider that document before blithely giving the force of law to any other agency "interpretations" of federal statutes.

Justice KAGAN, with whom Justice GINSBURG, Justice BREYER, and Justice SOTOMAYOR join, dissenting.

The Environmental Protection Agency placed emissions limits on coal and oil power plants following a lengthy regulatory process during which the Agency carefully considered costs. At the outset, EPA determined that regulating plants' emissions of hazardous air pollutants is "appropriate and necessary" given the harm they cause, and explained that it would take costs into account in developing suitable emissions standards. Next, EPA divided power plants into groups based on technological and other characteristics bearing significantly on their cost structures. It required plants in each group to match the emissions levels already achieved by the best-performing members of the same group — benchmarks necessarily reflecting those plants' own cost analyses. EPA then adopted a host of measures designed to make compliance with its proposed emissions limits less costly for plants that needed to catch up with their cleaner peers. And with only one narrow exception, EPA decided not to impose any more stringent standards (beyond what some plants had already achieved on their own) because it found that doing so would not be cost-effective. After all that, EPA conducted a formal cost-benefit study which found that the quantifiable benefits of its regulation would exceed the costs up to nine times over — by as much as $80 billion each year. Those benefits include as many as 11,000 fewer premature deaths annually, along with a far greater number of avoided illnesses.

Despite that exhaustive consideration of costs, the Court strikes down EPA's rule on the ground that the Agency "unreasonably... deemed cost irrelevant." Ante, at 2712. On the majority's theory, the rule is invalid because EPA did not explicitly analyze costs at the very first stage of the regulatory process, when making its "appropriate and necessary" finding. And that is so even though EPA later took costs into account again and again and ... so on. The majority thinks entirely immaterial, and so entirely ignores, all the subsequent times and ways EPA considered costs in deciding what any regulation would look like.

That is a peculiarly blinkered way for a court to assess the lawfulness of an agency's rulemaking. I agree with the majority — let there be no doubt about this — that EPA's power plant regulation would be unreasonable if "[t]he Agency gave cost no thought at all." Ante, at 2706 (emphasis in original). But that is just not what happened here. Over more than a decade, EPA took costs into account at multiple stages and through multiple means as it set emissions limits for power plants. And when making its initial "appropriate and necessary" finding, EPA knew it would do exactly that — knew it would thoroughly consider the cost-effectiveness of emissions standards later on. That context matters. The Agency acted well within its authority in declining to consider costs at the opening bell of the regulatory process given that it would do so in every round thereafter — and given that the emissions limits finally issued would depend crucially on those accountings. Indeed, EPA could not have measured costs at the process's initial stage with any accuracy. And the regulatory path EPA chose parallels the one it has trod in setting emissions limits, at Congress's explicit direction, for every other source of hazardous air pollutants over two decades. The majority's decision that EPA cannot take the same approach here — its micromanagement of EPA's rulemaking, based on little more than the word "appropriate" — runs counter to Congress's allocation of authority between the Agency and the courts. Because EPA reasonably found that it was "appropriate" to decline to analyze costs at a single stage of a regulatory proceeding otherwise imbued with cost concerns, I respectfully dissent.

I

A

The Clean Air Act Amendments of 1990, as the majority describes, obligate EPA to regulate emissions of mercury and other hazardous air pollutants from stationary sources discharging those substances in large quantities. See ante, at 2704-2705. For most industries, the statute prescribes the same multi-step regulatory process. At the initial stage, EPA must decide whether to regulate a source, based solely on the quantity of pollutants it emits and their health and environmental effects. See 42 U.S.C. §§ 7412(a)(1), (a)(2), (c)(1), (c)(3); ante, at 2704-2705. Costs enter the equation after that, affecting the emissions limits that the eventual regulation will require. Under the statute, EPA must divide sources into categories and subcategories and then set "floor standards" that reflect the average emissions level already achieved by the best-performing 12% of sources within each group. See § 7412(d)(3); ante, at 2705. Every 12% floor has cost concerns built right into it because the top sources, as successful actors in a market economy, have had to consider costs in choosing their own emissions levels. Moreover, in establishing categories and subcategories at this first stage, EPA can (significantly) raise or lower the costs of regulation for each source, because different classification schemes will alter the group — and so the emissions level — that the source has to match.1 Once the floor is set, EPA has to decide whether to impose any stricter ("beyond-the-floor") standards, "taking into consideration," among other things, "the cost of achieving such emissions reduction." § 7412(d)(2); see ante, at 2705. Finally, by virtue of a longstanding Executive Order applying to significant rules issued under the Clean Air Act (as well as other statutes), the Agency must systematically assess the regulation's costs and benefits. See Exec. Order No. 12866, 58 Fed.Reg. 51735, 51738, 51741 (1993) (applying to all rules with an annual economic effect of at least $100 million).

Congress modified that regulatory scheme for power plants. It did so because the 1990 amendments established a separate program to control power plant emissions contributing to acid rain, and many thought that just by complying with those requirements, plants might reduce their emissions of hazardous air pollutants to acceptable levels. See ante, at 2704-2705. That prospect counseled a "wait and see" approach, under which EPA would give the Act's acid rain provisions a chance to achieve that side benefit before imposing any further regulation. Accordingly, Congress instructed EPA to "perform a study of the hazards to public health reasonably anticipated" to result from power plants' emissions after the 1990 amendments had taken effect. § 7412(n)(1)(A). And Congress provided that EPA "shall regulate" those emissions only if the Agency "finds such regulation is appropriate and necessary after considering the results of the [public health] study." Ibid. Upon making such a finding, however, EPA is to regulate power plants as it does every other stationary source: first, by categorizing plants and setting floor standards for the different groups; then by deciding whether to regulate beyond the floors; and finally, by conducting the cost-benefit analysis required by Executive Order.

EPA completed the mandated health study in 1998, and the results gave much cause for concern. The Agency concluded that implementation of the acid rain provisions had failed to curb power plants' emissions of hazardous air pollutants. Indeed, EPA found, coal plants were on track to increase those emissions by as much as 30% over the next decade. See 1 EPA, Study of Hazardous Air Pollutant Emissions from Electric Utility Steam Generating Units — Final Report to Congress, p. ES-25 (1998). And EPA determined, focusing especially on mercury, that the substances released from power plants cause substantial health harms. Noting that those plants are "the largest [non-natural] source of mercury emissions," id., § 1.2.5.1, at 1-7, EPA found that children of mothers exposed to high doses of mercury during pregnancy "have exhibited a variety of developmental neurological abnormalities," including delayed walking and talking, altered muscles, and cerebral palsy. Id., § 7.2.2, at 7-17 to 7-18; see also 7 EPA, Mercury Study Report to Congress, p. 6-31 (1997) (Mercury Study) (estimating that 7% of women of childbearing age are exposed to mercury in amounts exceeding a safe level).

Informed by its public health study and additional data, EPA found in 2000 that it is "appropriate and necessary" to regulate power plants' emissions of mercury and other hazardous air pollutants. 65 Fed. Reg. 79830.2 Pulling apart those two adjectives, the Agency first stated that such regulation is "appropriate" because those pollutants "present[] significant hazards to public health and the environment" and because "a number of control options" can "effectively reduce" their emission. Ibid. EPA then determined that regulation is "necessary" because other parts of the 1990 amendments — most notably, the acid rain provisions — "will not adequately address" those hazards. Ibid. In less bureaucratic terms, EPA decided that it made sense to kick off the regulatory process given that power plants' emissions pose a serious health problem, that solutions to the problem are available, and that the problem will remain unless action is taken.

B

If the regulatory process ended as well as started there, I would agree with the majority's conclusion that EPA failed to adequately consider costs. Cost is almost always a relevant — and usually, a highly important — factor in regulation. Unless Congress provides otherwise, an agency acts unreasonably in establishing "a standard-setting process that ignore[s] economic considerations." Industrial Union Dept., AFL-CIO v. American Petroleum Institute, 448 U.S. 607, 670, 100 S.Ct. 2844, 65 L.Ed.2d 1010 (1980) (Powell, J., concurring in part and concurring in judgment). At a minimum, that is because such a process would "threaten[] to impose massive costs far in excess of any benefit." Entergy Corp. v. Riverkeeper, Inc., 556 U.S. 208, 234, 129 S.Ct. 1498, 173 L.Ed.2d 369 (2009) (BREYER, J., concurring in part and dissenting in part). And accounting for costs is particularly important "in an age of limited resources available to deal with grave environmental problems, where too much wasteful expenditure devoted to one problem may well mean considerably fewer resources available to deal effectively with other (perhaps more serious) problems." Id., at 233, 129 S.Ct. 1498; see ante, at 2707. As the Court notes, that does not require an agency to conduct a formal cost-benefit analysis of every administrative action. See ante, at 2711. But (absent contrary indication from Congress) an agency must take costs into account in some manner before imposing significant regulatory burdens.

That proposition, however, does not decide the issue before us because the "appropriate and necessary" finding was only the beginning. At that stage, EPA knew that a lengthy rulemaking process lay ahead of it; the determination of emissions limits was still years away. And the Agency, in making its kick-off finding, explicitly noted that consideration of costs would follow: "As a part of developing a regulation" that would impose those limits, "the effectiveness and costs of controls will be examined." 65 Fed.Reg. 79830. Likewise, EPA explained that, in the course of writing its regulation, it would explore regulatory approaches "allowing for least-cost solutions." Id., at 79830-79831. That means the Agency, when making its "appropriate and necessary" finding, did not decline to consider costs as part of the regulatory process. Rather, it declined to consider costs at a single stage of that process, knowing that they would come in later on.

The only issue in these cases, then, is whether EPA acted reasonably in structuring its regulatory process in that way — in making its "appropriate and necessary finding" based on pollution's harmful effects and channeling cost considerations to phases of the rulemaking in which emission levels are actually set. Said otherwise, the question is not whether EPA can reasonably find it "appropriate" to regulate without thinking about costs, full stop. It cannot, and it did not. Rather, the question is whether EPA can reasonably find it "appropriate" to trigger the regulatory process based on harms (and technological feasibility) alone, given that costs will come into play, in multiple ways and at multiple stages, before any emission limit goes into effect.

In considering that question, the very nature of the word "appropriate" matters. "[T]he word `appropriate,'" this Court has recognized, "is inherently context-dependent": Giving it content requires paying attention to the surrounding circumstances. Sossamon v. Texas, 563 U.S. 277, ___, 131 S.Ct. 1651, 1659, 179 L.Ed.2d 700 (2011). (That is true, too, of the word "necessary," although the majority spends less time on it. See Armour & Co. v. Wantock, 323 U.S. 126, 129-130, 65 S.Ct. 165, 89 S.Ct. 118 (1944) ("[T]he word `necessary' ... has always been recognized as a word to be harmonized with its context").) And here that means considering the place of the "appropriate and necessary" finding in the broader regulatory scheme — as a triggering mechanism that gets a complex rulemaking going. The interpretive task is thus at odds with the majority's insistence on staring fixedly "at this stage." Ante, at 2709 (emphasis in original). The task instead demands taking account of the entire regulatory process in thinking about what is "appropriate" in its first phase. The statutory language, in other words, is a directive to remove one's blinders and view things whole — to consider what it is fitting to do at the threshold stage given what will happen at every other.

And that instruction is primarily given to EPA, not to courts: Judges may interfere only if the Agency's way of ordering its regulatory process is unreasonable — i.e., something Congress would never have allowed. The question here, as in our seminal case directing courts to defer to agency interpretations of their own statutes, arises "not in a sterile textual vacuum, but in the context of implementing policy decisions in a technical and complex arena." Chevron U.S.A. Inc. v. Natural Resources Defense Council, Inc., 467 U.S. 837, 863, 104 S.Ct. 2778, 81 L.Ed.2d 694 (1984). EPA's experience and expertise in that arena — and courts' lack of those attributes — demand that judicial review proceed with caution and care. The majority actually phrases this principle well, though honors it only in the breach: Within wide bounds, it is "up to the Agency to decide... how to account for cost." Ante, at 2711. That judges might have made different regulatory choices — might have considered costs in different ways at different times — will not suffice to overturn EPA's action where Congress, as here, chose not to speak directly to those matters, but to leave them to the Agency to decide.

All of that means our decision here properly rests on something the majority thinks irrelevant: an understanding of the full regulatory process relating to power plants and of EPA's reasons for considering costs only after making its initial "appropriate and necessary" finding. I therefore turn to those issues, to demonstrate the simple point that should resolve these cases: that EPA, in regulating power plants' emissions of hazardous air pollutants, accounted for costs in a reasonable way.

II

A

In the years after its "appropriate and necessary" finding, EPA made good on its promise to account for costs "[a]s a part of developing a regulation." 65 Fed.Reg. 79830; see supra, at 2717. For more than a decade, as EPA deliberated on and then set emissions limits, costs came into the calculus at nearly every turn. Reflecting that consideration, EPA's final rule noted that steps taken during the regulatory process had focused on "flexib[ility] and cost-effective[ness]" and had succeeded in making "the rule less costly and compliance more readily manageable." 77 Fed.Reg. 9306, 9376. And the regulation concluded that "the benefits of th[e] rule" to public health and the environment "far outweigh the costs." Id., at 9306.

Consistent with the statutory framework, EPA initially calculated floor standards: emissions levels of the best-performing 12% of power plants in a given category or subcategory. The majority misperceives this part of the rulemaking process. It insists that EPA "must promulgate certain ... floor standards no matter the cost." Ante, at 2709. But that ignores two crucial features of the top-12% limits: first, the way in which any such standard intrinsically accounts for costs, and second, the way in which the Agency's categorization decisions yield different standards for plants with different cost structures.

The initial point is a fact of life in a market economy: Costs necessarily play a role in any standard that uses power plants' existing emissions levels as a benchmark. After all, the best-performing 12% of power plants must have considered costs in arriving at their emissions out-puts; that is how profit-seeking enterprises make decisions. And in doing so, they must have selected achievable levels; else, they would have gone out of business. (The same would be true even if other regulations influenced some of those choices, as the majority casually speculates. See ante, at 2710.) Indeed, this automatic accounting for costs is why Congress adopted a market-leader-based standard. As the Senate Report accompanying the 1990 amendments explained: "Cost considerations are reflected in the selection of emissions limitations which have been achieved in practice (rather than those which are merely theoretical) by sources of a similar type or character." S. Rep. No. 101-228, pp. 168-169 (1989), 1990 U.S.C.C.A.N. 3385, 3554. Of course, such a standard remains technology-forcing: It requires laggards in the industry to catch up with frontrunners, sometimes at significant expense. But the benchmark is, by definition, one that some power plants have achieved economically. And when EPA made its "appropriate and necessary" finding, it knew that fact — knew that the consequence of doing so was to generate floor standards with cost considerations baked right in.

Still more, EPA recognized that in making categorization decisions, it could take account of multiple factors related to costs of compliance — and so avoid impracticable regulatory burdens. Suppose, to use a simple example, that curbing emissions is more technologically difficult — and therefore more costly — for plants burning coal than for plants burning oil. EPA can then place those two types of plants in different categories, so that coal plants need only match other coal plants rather than having to incur the added costs of meeting the top oil plants' levels. Now multiply and complexify that example many times over. As the Agency noted when making its "appropriate and necessary" finding, EPA "build[s] flexibility" into the regulatory regime by "bas[ing] subcategorization on ... the size of a facility; the type of fuel used at the facility; and the plant type," and also "may consider other relevant factors such as geographic conditions." 65 Fed. Reg. 79830; see S.Rep. No. 101-228, at 166 (listing similar factors and noting that "[t]he proper definition of categories ... will assure maximum protection of public health and the environment while minimizing costs imposed on the regulated community"). Using that classification tool, EPA can ensure that plants have to attain only the emissions levels previously achieved by peers facing comparable cost constraints, so as to further protect plants from unrealistic floor standards.

And that is exactly what EPA did over the course of its rulemaking process, insisting on apples-to-apples comparisons that bring floor standards within reach of diverse kinds of power plants. Even in making its "appropriate and necessary" finding, the Agency announced it would divide plants into the two categories mentioned above: "coal-fired" and "oil-fired." 65 Fed.Reg. 79830.3 Then, as the rulemaking progressed, EPA went further. Noting that different technologies significantly affect the ease of attaining a given emissions level, the Agency's proposed rule subdivided those two classes into five: plants designed to burn high-rank coal; plants designed to burn low-rank virgin coal; plants that run on a technology termed integrated gasification combined cycle; liquid oil units; and solid oil units. See 76 Fed.Reg. 25036-25037. EPA explained that by subcategorizing in that way, it had spared many plants the need to "retrofit[]," "redesign[]," or make other "extensive changes" to their facilities. Id., at 25036. And in its final rule, EPA further refined its groupings in ways that eased compliance. Most notably, the Agency established a separate subcategory, and attendant (less stringent) floor, for plants in Hawaii, Puerto Rico, Guam, and the Virgin Islands on the ground that plants in those places have "minimal control over the quality of available fuel[] and disproportionately high operational and maintenance costs." 77 Fed.Reg. 9401.4

Even after establishing multiple floor standards that factored in costs, EPA adopted additional "compliance options" to "minimize costs" associated with attaining a given floor — just as its "appropriate and necessary" finding explicitly contemplated. Id., at 9306; 76 Fed.Reg. 25057; see 65 Fed.Reg. 79830. For example, the Agency calculated each floor as both an "input-based" standard (based on emissions per unit of energy used) and an "output-based" standard (based on emissions per unit of useful energy produced), and allowed plants to choose which standard they would meet. That option, EPA explained, can "result in ... reduced compliance costs." 76 Fed.Reg. 25063. Similarly, EPA allowed plants to meet a given 12% floor by averaging emissions across all units at the same site, instead of having to meet the floor at each unit. Some plants, EPA understood, would find such averaging a "less costly alternative." 77 Fed.Reg. 9385. Yet again: EPA permitted "limited use" plants — those primarily burning natural gas but sometimes switching to oil — to comply with the final rule by meeting qualitative "work practice standards" rather than numeric emissions limits. Id., at 9400-9401. EPA explained that it would be "economically impracticable" for those plants to demonstrate compliance through emissions testing, and that an alternative standard, focused on their adoption of pollution control techniques, would allow them to both reduce emissions and avoid "extra cost." Id., at 9401. And the list goes on. See, e.g., id., at 9409-9410 (allowing extra year for plants to comply with emissions limits where "source-specific construction, permitting, or labor, procurement or resource challenges" arise); id., at 9417 (describing additional "compliance options").

With all that cost-consideration under its belt, EPA next assessed whether to set beyond-the-floor standards, and here too, as it knew it would, the Agency took costs into account. For the vast majority of coal and oil plants, EPA decided that beyond-the-floor standards would not be "reasonable after considering costs." Id., at 9331. The Agency set such a standard for only a single kind of plant, and only after determining that the technology needed to meet the more lenient limit would also achieve the more stringent one. See id., at 9393; 76 Fed.Reg. 25046-25047. Otherwise, EPA determined, the market-leader-based standards were enough.

Finally, as required by Executive Order and as anticipated at the time of the "appropriate and necessary" finding, EPA conducted a formal cost-benefit analysis of its new emissions standards and incorporated those findings into its proposed and final rules. See id., at 25072-25078; 77 Fed.Reg. 9305-9306, 9424-9432. That analysis estimated that the regulation's yearly costs would come in at under $10 billion, while its annual measureable benefits would total many times more — between $37 and $90 billion. See id., at 9305-9306; ante, at 2705-2706. On the costs side, EPA acknowledged that plants' compliance with the rule would likely cause electricity prices to rise by about 3%, but projected that those prices would remain lower than they had been as recently as 2010. See 77 Fed.Reg. 9413-9414. EPA also thought the rule's impact on jobs would be about a wash, with jobs lost at some high-emitting plants but gained both at cleaner plants and in the pollution control industry. See ibid. On the benefits side, EPA noted that it could not quantify many of the health gains that would result from reduced mercury exposure. See id., at 9306. But even putting those aside, the rule's annual benefits would include between 4,200 and 11,000 fewer premature deaths from respiratory and cardiovascular causes, 3,100 fewer emergency room visits for asthmatic children, 4,700 fewer non-fatal heart attacks, and 540,000 fewer days of lost work. See id., at 9429.

Those concrete findings matter to these cases — which, after all, turn on whether EPA reasonably took costs into account in regulating plants' emissions of hazardous air pollutants. The majority insists that it may ignore EPA's cost-benefit analysis because "EPA did not rely on" it when issuing the initial "appropriate and necessary" finding. Ante, at 2711 (quoting Solicitor General); see also SEC v. Chenery Corp., 318 U.S. 80, 87, 93-94, 63 S.Ct. 454, 87 S.Ct. 626 (1943). At one level, that description is true — indeed, a simple function of chronology: The kick-off finding preceded the cost-benefit analysis by years and so could not have taken its conclusions into account. But more fundamentally, the majority's account is off, because EPA knew when it made that finding that it would consider costs at every subsequent stage, culminating in a formal cost-benefit study. And EPA knew that, absent unusual circumstances, the rule would need to pass that cost-benefit review in order to issue. See Exec. Order No. 12866, 58 Fed.Reg. 51736 ("Each agency shall ... adopt a regulation only upon a reasoned determination that the benefits of the intended regulation justify its costs"). The reasonableness of the Agency's decision to consider only the harms of emissions at the threshold stage must be evaluated in that broader context. And in thinking about that issue, it is well to remember the out-come here: a rule whose benefits exceed its costs by three to nine times. In making its "appropriate and necessary" finding, EPA had committed to assessing and mitigating costs throughout the rest of its rulemaking; if nothing else, the findings of the Agency's cost-benefit analysis — making clear that the final emissions standards were cost-effective — show that EPA did just that.

B

Suppose you were in charge of designing a regulatory process. The subject matter — an industry's emissions of hazardous material — was highly complex, involving multivarious factors demanding years of study. Would you necessarily try to do everything at once? Or might you try to break down this lengthy and complicated process into discrete stages? And might you consider different factors, in different ways, at each of those junctures? I think you might. You know that everything must get done in the end — every relevant factor considered. But you tend to think that "in the end" does not mean "in the beginning." And you structure your rulemaking process accordingly, starting with a threshold determination that does not mirror your end-stage analysis. Would that be at least (which is all it must be) a "reasonable policy choice"? Chevron, 467 U.S., at 845, 104 S.Ct. 2778.

That is the question presented here, and it nearly answers itself. Setting emissions levels for hazardous air pollutants is necessarily a lengthy and complicated process, demanding analysis of many considerations over many years. Costs are a key factor in that process: As I have said, sensible regulation requires careful scrutiny of the burdens that potential rules impose. See supra, at 2716-2717. But in ordering its regulatory process, EPA knew it would have the opportunity to consider costs in one after another of that rulemaking's stages — in setting the level of floor standards, in providing a range of options for plants to meet them, in deciding whether or where to require limits beyond the floor, and in finally completing a formal cost-benefit analysis. See 65 Fed.Reg. 79830-79831; supra, at 2718-2721. Given that context, EPA reasonably decided that it was "appropriate" — once again, the only statutory requirement relevant here — to trigger the regulatory process based on the twin findings that the emissions in question cause profound health and environmental harms and that available pollution control technologies can reduce those emissions. By making that decision, EPA did no more than commit itself to developing a realistic and cost-effective regulation — a rule that would take account of every relevant factor, costs and benefits alike. And indeed, particular features of the statutory scheme here indicate that EPA's policy choice was not just a minimally reasonable option but an eminently reasonable one.

To start, that decision brought EPA's regulation of power plants into sync with its regulation of every other significant source of hazardous pollutants under the Clean Air Act. For all those types of sources (totaling over 100), the Act instructs EPA to make the threshold decision to regulate based solely on the quantity and effects of pollutants discharged; costs enter the picture afterward, when the Agency takes up the task of actually establishing emissions limits. See supra, at 2715. Industry after industry, year after year, EPA has followed that approach to standard-setting, just as Congress contemplated. See, e.g., 58 Fed.Reg. 49354 (1993) (dry cleaning facilities); 59 Fed. Reg. 64303 (1994) (gasoline distributors); 60 Fed.Reg. 45948 (1995) (aerospace manufacturers). And apparently with considerable success. At any rate, neither those challenging this rule nor the Court remotely suggests that these regulatory regimes have done "significantly more harm than good." Ante, at 2707. So when making its "appropriate and necessary" finding for power plants, EPA had good reason to continue in the same vein. See, e.g., Entergy, 556 U.S., at 236, 129 S.Ct. 1498 (opinion of BREYER, J.) (noting that the reasonableness of an agency's approach to considering costs rests in part on whether that tack has met "with apparent success in the past"). And that is exactly how EPA explained its choice. Stating that it would consider the "costs of controls" when "developing a regulation," the Agency noted that such an "approach has helped build flexibility in meeting environmental objectives in the past," thereby preventing the imposition of disproportionate costs. 65 Fed.Reg. 79830. Indeed, as EPA further commented in issuing its rule, it would seem "inequitable to impose a regulatory regime on every industry in America and then to exempt one category" after finding it represented "a significant part of the air toxics problem." 77 Fed. Reg. 9322 (quoting 136 Cong. Rec. 36062 (1990) (statement of Sen. Durenberger)).

The majority's attempt to answer this point founders on even its own statement of facts. The majority objects that "the whole point of having a separate provision about power plants" is to "treat[] power plants differently from other stationary sources." Ante, at 2710 (emphasis in original). But turn back about 10 pages, and read what the majority says about why Congress treated power plants differently: because, as all parties agree, separate regulatory requirements involving acid rain "were expected to have the collateral effect of reducing power plants' emissions of hazardous air pollutants, although the extent of the reduction was unclear." Ante, at 2705; see supra, at 2715. For that reason alone (the majority does not offer any other), Congress diverted EPA from its usual regulatory path, instructing the Agency, as a preliminary matter, to complete and consider a study about the residual harms to public health arising from those emissions. See ante, at 2704; supra, at 2716. But once EPA found in its study that the acid rain provisions would not significantly affect power plants' emissions of hazardous pollutants, any rationale for treating power plants differently from other sources discharging the same substances went up in smoke. See 65 Fed. Reg. 79830. At that point, the Agency would have had far more explaining to do if, rather than following a well-tested model, it had devised a new scheme of regulation for power plants only.

Still more, EPA could not have accurately assessed costs at the time of its "appropriate and necessary" finding. See 8 Mercury Study, at 6-2 (noting the "many uncertainties" in any early-stage analysis of pollution control costs). Under the statutory scheme, that finding comes before — years before — the Agency designs emissions standards. And until EPA knows what standards it will establish, it cannot know what costs they will impose. Nor can those standards even be reasonably guesstimated at such an early stage. Consider what it takes to set floor standards alone. First, EPA must divide power plants into categories and subcategories; as explained earlier, those classification decisions significantly affect what floors are established. See supra, at 2715, and n. 1, 2720-2721. And then, EPA must figure out the average emissions level already achieved by the top 12% in each class so as to set the new standards. None of that can realistically be accomplished in advance of the Agency's regulatory process: Indeed, those steps are the very stuff of the rulemaking. Similarly, until EPA knows what "compliance options" it will develop, it cannot know how they will mitigate the costs plants must incur to meet the floor standards. See supra, at 2720-2721. And again, deciding on those options takes substantial time. So there is good reason for different considerations to go into the threshold finding than into the final rule. Simply put, calculating costs before starting to write a regulation would put the cart before the horse.

III

The central flaw of the majority opinion is that it ignores everything but one thing EPA did. It forgets that EPA's "appropriate and necessary" finding was only a first step which got the rest of the regulatory process rolling. It narrows its field of vision to that finding in isolation, with barely a glance at all the ways in which EPA later took costs into account. See supra, at 2718-2719 (in establishing floor standards); supra, at 2720-2721 (in adopting compliance options); supra, at 2722 (in deciding whether to regulate beyond the floor); supra, at 2721-2722 (in conducting a formal cost-benefit analysis as a final check). In sum, the majority disregards how consideration of costs infused the regulatory process, resulting not only in EPA's adoption of mitigation measures, ante, at 2711, but also in EPA's crafting of emissions standards that succeed in producing benefits many times their price.

That mistake accounts for the majority's primary argument that the word "appropriate," as used in § 7412(n)(1)(A), demands consideration of costs. See ante, at 2707-2708. As I have noted, that would be true if the "appropriate and necessary" finding were the only step before imposing regulations on power plants. See supra, at 2716-2717. But, as should be more than clear by now, it was just the first of many: Under the Clean Air Act, a long road lay ahead in which the Agency would have more — and far better — opportunities to evaluate the costs of diverse emissions standards on power plants, just as it did on all other sources. See supra, at 2715, 2717, 2718-2721. EPA well understood that fact: "We evaluate the terms `appropriate' and `necessary,'" it explained, in light of their "statutory context." 76 Fed. Reg. 24986. And EPA structured its regulatory process accordingly, with consideration of costs coming (multiple times) after the threshold finding. The only way the majority can cast that choice as unreasonable, given the deference this Court owes to such agency decisions, is to blind itself to the broader rulemaking scheme.

The same fault inheres in the majority's secondary argument that EPA engaged in an "interpretive gerrymander[]" by considering environmental effects but not costs in making its "appropriate and necessary" finding. Ante, at 2708. The majority notes — quite rightly — that Congress called for EPA to examine both subjects in a study of mercury emissions from all sources (separate from the study relating to power plants' emissions alone). See ante, at 2708. And the majority states — again, rightly — that Congress's demand for that study "provides direct evidence that Congress was concerned with [both] environmental effects [and] cost." Ante, at 2708 (internal quotation marks omitted). But nothing follows from that fact, because EPA too was concerned with both. True enough, EPA assessed the two at different times: environmental harms (along with health harms) at the threshold, costs after-ward. But that was for the very reasons earlier described: because EPA wanted to treat power plants like other sources and because it thought harms, but not costs, could be accurately measured at that early stage. See supra, at 2722-2724. Congress's simple request for a study of mercury emissions in no way conflicts with that choice of when and how to consider both harms and costs. Once more, the majority perceives a conflict only because it takes so partial a view of the regulatory process.

And the identical blind spot causes the majority's sports-car metaphor to run off the road. The majority likens EPA to a hypothetical driver who decides that "it is `appropriate' to buy a Ferrari without thinking about cost, because he plans to think about cost later when deciding whether to upgrade the sound system." Ante, at 2709. The comparison is witty but wholly inapt. To begin with, emissions limits are not a luxury good: They are a safety measure, designed to curtail the significant health and environmental harms caused by power plants spewing hazardous pollutants. And more: EPA knows from past experience and expertise alike that it will have the opportunity to purchase that good in a cost-effective way. A better analogy might be to a car owner who decides without first checking prices that it is "appropriate and necessary" to replace her worn-out brake-pads, aware from prior experience that she has ample time to comparison-shop and bring that purchase within her budget. Faced with a serious hazard and an available remedy, EPA moved forward like that sensible car owner, with a promise that it would, and well-grounded confidence that it could, take costs into account down the line.

That about does it for the majority's opinion, save for its final appeal to Chenery — and Chenery cannot save its holding. See ante, at 2709-2710. Of course a court may not uphold agency action on grounds different from those the agency gave. See Chenery, 318 U.S., at 87, 63 S.Ct. 454. But equally, a court may not strike down agency action without considering the reasons the agency gave. Id., at 95, 63 S.Ct. 454. And that is what the majority does. Indeed, it is difficult to know what agency document the majority is reading. It denies that "EPA said ... that cost-benefit analysis would be deferred until later." Ante, at 2710. But EPA said exactly that: The "costs of controls," the Agency promised, "will be examined" as "a part of developing a regulation." 65 Fed.Reg. 79830. Tellingly, these words appear nowhere in the majority's opinion. But what are they other than a statement that cost concerns, contra the majority, are not "irrelevant," ante, at 2710 (without citation) — that they are simply going to come in later?

And for good measure, EPA added still extra explanation. In its "appropriate and necessary" finding, the Agency committed to exploring "least-cost solutions" in "developing a standard for utilities." 65 Fed. Reg. 79830. The Agency explained that such an approach — particularly mentioning the use of averaging and subcategorization — had offered "opportunit[ies] for lower cost solutions" and "helped build flexibility in meeting environmental objectives in the past." Ibid.; see supra, at 2717, 2723. Then, in issuing its proposed and final rules, EPA affirmed that it had done just what it said. EPA recognized that standard-setting must "allow the industry to make practical investment decisions that minimize costs." 76 Fed.Reg. 25057. Accordingly, the Agency said, it had "provid[ed] flexibility and compliance options" so as to make the rule "less costly" for regulated parties. 77 Fed.Reg. 9306. EPA added that it had rejected beyond-the-floor standards for almost all power plants because they would not be "reasonable after considering costs." Id., at 9331. And it showed the results of a formal analysis finding that the rule's costs paled in comparison to its benefits. In sum, EPA concluded, it had made the final standards "cost-efficient." Id., at 9434. What more would the majority have EPA say?

IV

Costs matter in regulation. But when Congress does not say how to take costs into account, agencies have broad discretion to make that judgment. Accord, ante, at 2711 (noting that it is "up to the Agency to decide (as always, within the limits of reasonable interpretation) how to account for cost"). Far more than courts, agencies have the expertise and experience necessary to design regulatory processes suited to "a technical and complex arena." Chevron, 467 U.S., at 863, 104 S.Ct. 2778. And in any event, Congress has entrusted such matters to them, not to us.

EPA exercised that authority reasonably and responsibly in setting emissions standards for power plants. The Agency treated those plants just as it had more than 100 other industrial sources of hazardous air pollutants, at Congress's direction and with significant success. It made a threshold finding that regulation was "appropriate and necessary" based on the harm caused by power plants' emissions and the availability of technology to reduce them. In making that finding, EPA knew that when it decided what a regulation would look like — what emissions standards the rule would actually set — the Agency would consider costs. Indeed, EPA expressly promised to do so. And it fulfilled that promise. The Agency took account of costs in setting floor standards as well as in thinking about beyond-the-floor standards. It used its full kit of tools to minimize the expense of complying with its proposed emissions limits. It capped the regulatory process with a formal analysis demonstrating that the benefits of its rule would exceed the costs many times over. In sum, EPA considered costs all over the regulatory process, except in making its threshold finding — when it could not have measured them accurately anyway. That approach is wholly consonant with the statutory scheme. Its adoption was "up to the Agency to decide." Ante, at 2711.

The majority arrives at a different conclusion only by disregarding most of EPA's regulatory process. It insists that EPA must consider costs — when EPA did just that, over and over and over again. It concedes the importance of "context" in determining what the "appropriate and necessary" standard means, see ante, at 2707-2708, 2709 — and then ignores every aspect of the rulemaking context in which that standard plays a part. The result is a decision that deprives the Agency of the latitude Congress gave it to design an emissions-setting process sensibly accounting for costs and benefits alike. And the result is a decision that deprives the American public of the pollution control measures that the responsible Agency, acting well within its delegated authority, found would save many, many lives. I respectfully dissent.