concurring in part and dissenting in part.
I join Parts I and II of the Court’s opinion; I agree that these eases are not moot and that the District Court had jurisdiction. I do not join Part III. The Court there gives effect to a reading of the Environmental Protection Agency’s regulations that is not the most natural one, simply because EPA says that it believes the unnatural reading is right. It does this, moreover, even though the Agency has vividly illustrated that it can write a rule saying precisely what it means—by doing just that while these cases were being briefed.
Enough is enough.
I
For decades, and for no good reason, we have been giving agencies the authority to say what their rules mean, under the harmless-sounding banner of “defer[ring] to an agency’s interpretation of its own regulations.” Talk America, Inc. v. Michigan Bell Telephone Co., 564 U. S. 50, 67 (2011) (Scalia, J., concurring). This is generally called Seminole Rock or Auer deference. See Bowles v. Seminole Rock & *617Sand Co., 325 U. S. 410 (1945); Auer v. Robbins, 519 U. S. 452 (1997).
Two Terms ago, in my separate concurrence in Talk America, I expressed doubts about the validity of this practice. In that case, however, the agency’s interpretation of the rule was also the fairest one, and no party had asked us to reconsider Auer. Today, however, the Court’s deference to the Agency makes the difference (note the Court’s defensive insistence that the Agency’s interpretation need not be “the best one,” ante, at 613). And respondent has asked us, if necessary, to “ ‘reconsider Auer.’ ” I believe that it is time to do so. See Brief for Respondent 42, n. 12; see also Brief for Law Professors on the Propriety of Administrative Deference as Amici Curiae. This is especially true because the circumstances of these cases illustrate Auer’s, flaws in a particularly vivid way.
The canonical formulation of Auer deference is that we will enforce an agency’s interpretation of its own rules unless that interpretation is “plainly erroneous or inconsistent with the regulation.” Seminole Rock, supra, at 414. But of course whenever the agency’s interpretation of the regulation is different from the fairest reading, it is in that sense “inconsistent” with the regulation. Obviously, that is not enough, or there would be nothing for Auer to do. In practice, Auer deference is Chevron deference applied to regulations rather than statutes. See Chevron U. S. A. Inc. v. Natural Resources Defense Council, Inc., 467 U. S. 837 (1984). The agency’s interpretation will be accepted if, though not the fairest reading of the regulation, it is a plausible reading—within the scope of the ambiguity that the regulation contains.
Our cases have not put forward a persuasive justification for Auer deference. The first case to apply it, Seminole Rock, offered no justification whatever—just the ipse dixit that “the administrative interpretation . . . becomes of controlling weight unless it is plainly erroneous or inconsistent *618with the regulation.” 325 U. S., at 414. Our later cases provide two principal explanations, neither of which has much to be said for it. See generally Stephenson & Pogoriler, Seminole Rock’s Domain, 79 Geo. Wash. L. Rev. 1449, 1454-1458 (2011). First, some cases say that the agency, as the drafter of the rule, will have some special insight into its intent when enacting it. E. g., Martin v. Occupational Safety and Health Review Comm’n, 499 U. S. 144, 150-153 (1991). The implied premise of this argument—that what we are looking for is the agency’s intent in adopting the rule—is false. There is true of regulations what is true of statutes. As Justice Holmes put it: “We do not inquire what the legislature meant; we ask only what the statute means.” The Theory of Legal Interpretation, 12 Harv. L. Rev. 417, 419 (1899). Whether governing rules are made by the National Legislature or an administrative agency, we are bound by what they say, not by the unexpressed intention of those who made them.
The other rationale our cases provide is that the agency possesses special expertise in administering its “‘complex and highly technical regulatory program.’” See, e. g., Thomas Jefferson Univ. v. Shalala, 512 U. S. 504, 512 (1994). That is true enough, and it leads to the conclusion that agencies and not courts should make regulations. But it has nothing to do with who should interpret regulations—unless one believes that the purpose of interpretation is to make the regulatory program work in a fashion that the current leadership of the agency deems effective. Making regulatory programs effective is the purpose of rulemaking, in which the agency uses its “special expertise” to formulate the best rule. But the purpose of interpretation is to determine the fair meaning of the rule—to “say what the law is,” Marbury v. Madison, 1 Cranch 137, 177 (1803). Not to make policy, but to determine what policy has been made and promulgated by the agency, to which the public owes obedience. Indeed, since the leadership of agencies (and hence the policy *619preferences of agencies) changes with Presidential administrations, an agency head can only be sure that the application of his “special expertise” to the issue addressed by a regulation will be given effect if we adhere to predictable principles of textual interpretation rather than defer to the “special expertise” of his successors. If we take agency enactments as written, the Executive has a stable background against which to write its rules and achieve the policy ends it thinks best.
Another conceivable justification for Auer deference, though not one that is to be found in our cases, is this: If it is reasonable to defer to agencies regarding the meaning of statutes that Congress enacted, as we do per Chevron, it is a fortiori reasonable to defer to them regarding the meaning of regulations that they themselves crafted. To give an agency less control over the meaning of its own regulations than it has over the meaning of a congressionally enacted statute seems quite odd.
But it is not odd at all. The theory of Chevron (take it or leave it) is that when Congress gives an agency authority to administer a statute, including authority to issue interpretive regulations, it implicitly accords the agency a degree of discretion, which the courts must respect, regarding the meaning of the statute. See Smiley v. Citibank (South Dakota), N. A., 517 U. S. 735, 740-741 (1996). While the implication of an agency power to clarify the statute is reasonable enough, there is surely no congressional implication that the agency can resolve ambiguities in its own regulations. For that would violate a fundamental principle of separation of powers—that the power to write a law and the power to interpret it cannot rest in the same hands. “When the legislative and executive powers are united in the same person ... there can be no liberty; because apprehensions may arise, lest the same monarch or senate should enact tyrannical laws, to execute them in a tyrannical manner.” Montesquieu, Spirit of the Laws bk. XI, ch. 6, pp. 151-152 (O. Piest *620ed., T. Nugent transí. 1949). Congress cannot enlarge its own power through Chevron—whatever it leaves vague in the statute will be worked out by someone else. Chevron represents a presumption about who, as between the Executive and the Judiciary, that someone else will be. (The Executive, by the way—the competing political branch—is the less congenial repository of the power as far as Congress is concerned.) So Congress’s incentive is to speak as clearly as possible on the matters it regards as important.
But when an agency interprets its own rules—that is something else. Then the power to prescribe is augmented by the power to interpret; and the incentive is to speak vaguely and broadly, so as to retain a “flexibility” that will enable “clarification” with retroactive effect. “It is perfectly understandable” for an agency to “issue vague regulations” if doing so will “maximiz[e] agency power.” Thomas Jefferson Univ., supra, at 525 (Thomas, J., dissenting). Combining the power to prescribe with the power to interpret is not a new evil: Blackstone condemned the practice of resolving doubts about “the construction of the Roman laws” by “stat[ing] the case to the emperor in writing, and tak[ing] his opinion upon it.” 1 W. Blackstone, Commentaries on the Laws of England 58 (1765). And our Constitution did not mirror the British practice of using the House of Lords as a court of last resort, due in part to the fear that he who has “agency in passing bad laws” might operate in the “same spirit” in their interpretation. The Federalist No. 81, pp. 543-544 (J. Cooke ed. 1961). Auer deference encourages agencies to be “vague in framing regulations, with the plan of issuing ‘interpretations’ to create the intended new law without observance of notice and comment procedures.” Anthony, The Supreme Court and the APA: Sometimes They Just Don’t Get It, 10 Admin. L. J. Am. U. 1,11-12 (1996). Auer is not a logical corollary to Chevron but a dangerous permission slip for the arrogation of power. See Talk America, 564 U. S., at 68-69 (Scalia, J., concurring); Manning, Constitu*621tional Structure and Judicial Deference to Agency Interpretations of Agency Rules, 96 Colum. L. Rev. 612 (1996).
It is true enough that Auer deference has the same beneficial pragmatic effect as Chevron deference: The country need not endure the uncertainty produced by divergent views of numerous district courts and courts of appeals as to what is the fairest reading of the regulation, until a definitive answer is finally provided, years later, by this Court. The agency’s view can be relied upon, unless it is, so to speak, beyond the pale. But the duration of the uncertainty produced by a vague regulation need , not be as long as the uncertainty produced by a vague statute. For as soon as an interpretation uncongenial to the agency is pronounced by a district court, the agency can begin the process of amending the regulation to make its meaning entirely clear. The circumstances of this case demonstrate the point. While these cases were being briefed before us, EPA issued a rule designed to respond to the Court of Appeals judgment we are reviewing. See 77 Fed. Reg. 72974 (2012) (to be codified in 40 CFR pt. 122, subpt. B). It did so (by the standards of such things) relatively quickly: The decision below was handed down in May 2011, and in December 2012 EPA published an amended rule setting forth in unmistakable terms the position it argues here. And there is another respect in which a lack of Chevron-type deference has less severe pragmatic consequences for rules than for statutes. In many cases, when an agency believes that its rule permits conduct that the text arguably forbids, it can simply exercise its discretion not to prosecute. That is not possible, of course, when, as here, a party harmed by the violation has standing to compel enforcement.
In any case, however great may be the efficiency gains derived from Auer deference, beneficial effect cannot justify a rule that not only has no principled basis but contravenes one of the great rules of separation of powers: He who writes a law must not adjudge its violation.
*622p-H hH
I would therefore resolve these cases by using the familiar tools of textual interpretation to decide: Is what the petitioners did here proscribed by the fairest reading of the regulations? What they did was to channel stormwater runoff from logging roads without a permit. To decide whether that was permissible we must answer one, and possibly two, questions: First, was the stormwater discharged from a “point source”? If not, no permit was required. But if so, we face the second question: Were the stormwater discharges exempt from the permit requirement because they were not “associated with industrial activity”? The fairest reading of the statute and regulations is that these discharges were from point sources, and were associated with industrial activity.
A
The Clean Water Act generally prohibits discharging pollution without a permit from what it calls a “point source.” 33 U. S. C. § 1311(a). A “point source” is defined as “any discernible, confined and discrete conveyance, including but not limited to any pipe, ditch, channel, tunnel, conduit,” and several other things. § 1362(14). The stormwater here was discharged from logging roads through a series of pipes, ditches, and channels—all items expressly named in the definition.
EPA argues that the Silvicultural Rule, 40 CFR § 122.27(b)(1) (2006), excludes from the definition of “[silvi-cultural point source” “harvesting operations ... from which there is natural runoff.” This is relevant, says the Agency, because that rule specifies that only “[sjilvicultural point sources, as defined in this section,” are “point sources subject to the . . . permit program.” § 122.27(a). In EPA’s view, the stormwater here is “natural runoff.”
But are stormwater discharges “natural runoff” when they are channeled through manmade pipes and ditches, and *623carry with them manmade pollutants from manmade forest roads? It is not obvious that this is so—as the Agency agrees. See Brief for United States as Amicus Curiae 19 (the rule’s “reference to ‘natural runoff’ associated with logging roads neither clearly encompasses nor clearly excludes the sort of channeled runoff that is at issue in this case”). In my view, giving the term the Agency’s interpretation would contradict the statute’s definition of “point source,” which explicitly includes any “pipe, ditch, channel, tunnel, [and] conduit.” Applying the interpretive presumption of validity— the canon that we are to “prefe[r] the meaning that preserves to the meaning that destroys,” Panama Refining Co. v. Ryan, 293 U. S. 388, 439 (1935) (Cardozo, J., dissenting)—I would hold that the regulation’s exclusion of “natural runoff” does not reach the situation here. The stormwater discharges came from point sources, because they flowed out of artificial “pipe[s],” “ditch[es],” and “channels],” 33 U. S. C. § 1362(14), and were thus not “natural runoff” from a logging operation, 40 CFR § 122.27(b)(1) (emphasis added).
B
Many point-source stormwater discharges are nonetheless exempt from the usual permitting requirement. See 33 U. S. C. § 1342(p). This exemption, however, does not reach discharges “associated with industrial activity.” Ibid. EPA has enacted a rule defining what it means for storm-water discharges to be “associated with” industrial activity, and what activities count as “industrial.” 40 CFR § 122.26(b)(14).
The regulation sets out 11 “categories of industries”; as to those industries, discharges are “associated with industrial activity” if they come from sites used for “transportation” of “any raw material.” Ibid. The forest roads at issue here are used to transport raw material (logs); the only question is whether logging is a “categor[y] of industr[y]” enumerated *624in the definition. It is: The second of the listed “categories of facilities” is “[facilities classified as Standard Industrial Classifications 24 (except 2434).” § 122.26(b)(14)(ii). Opening one’s hymnal to Standard Industrial Classification 24 (“Lumber and Wood Products, Except Furniture”), one finds that the first industry group listed, No. 2411, is “Logging”— defined as “[establishments primarily engaged in cutting timber.” 2 App. 64. (As if that were not clear enough, an illustrative product of this industry is helpfully listed: “Logs.”) That, I would think, is that.
EPA disagrees, and the Court gives the Agency’s position Auer deference, but that reading is certainly not the most natural one. The Court relies heavily on the fact that the definition of “[s]torm water discharge associated with industrial activity” requires that the discharge be “directly related to manufacturing, processing or raw materials storage areas at an industrial plant,” § 122.26(b)(14). The crucial question this definition presents is whether the concluding phrase “at an industrial plant” limits only the last noun phrase (“raw materials storage areas”) or also the two preceding nouns (“manufacturing” and “processing”). The canon of interpretation known as the rule of the last antecedent states that “a limiting clause or phrase . . . should ordinarily be read as modifying only the noun or phrase that it immediately follows.” Barnhart v. Thomas, 540 U. S. 20, 26 (2003). If a statute provides that “it shall be unlawful to possess a grenade launcher, a fully-automatic weapon, or a shotgun with a barrel shorter than 12 inches,” that does not mean that a grenade launcher with a barrel longer than 12 inches is legal. Application of the canon would mean that “at an industrial plant” modifies only “raw materials storage areas,” and therefore that “manufacturing” and “processing” anywhere, including in the forest, would be “associated with industrial activity.” (Standard Industrial Classification 24 categorizes logging as a manufacturing business, and these discharges are therefore “directly related to manufacturing”)
*625Like all canons of interpretation, the rule of the last antecedent can be overcome by textual indication of contrary meaning. But that does not exist here. To the contrary, the. enumerated categories of industries to which the term “industrial activity” applies reinforce the proposition that “at an industrial plant” does not modify “manufacturing” or “processing.” The term includes (in addition to logging) “active or inactive mining operations,” § 122.26(b)(14)(iii); “[IJandfills” and “open dumps,” § 122.26(b)(14)(v); “automobile junkyards,” § 122.26(b)(14)(vi); and “Construction activity including clearing, grading and excavation,” § 122.26(b) (14)(x). Those industries and activities (while related to manufacturing and processing) virtually never take place at anything like what one might describe as a “plant.” The rule of the last antecedent is therefore confirmed as the correct guide to meaning here: “at an industrial plant” limits only “raw materials storage areas.”
EPA also insists, Brief for United States as Amicus Curiae 24, that the regulation reaches only “ ‘traditional’ ” sources of industrial stormwater, such as sawmills. But Standard Industrial Classification 24. has a specific subcategory (No. 242) that is “Sawmills and Planing Mills.” 2 App. 64. The rule is not so limited, reaching by its terms “Standard Industrial Classifieatio[n] 24 (except 2434).” § 122.26(b) (14)(ii). The explicit carving-out of No. 2434 is telling: Why EPA chose to exclude “establishments primarily engaged in manufacturing wood kitchen cabinet and wood bathroom vanities” from the definition of industrial stormwater, I do not know—but the picayune nature of the exclusion gives lie to the idea that the rule’s scope ought to be decided by a rough sense of its gestalt. If EPA had meant to reach only sawmills, it quite obviously knew how to do so.
Finally, the Court believes that Standard Industrial Classification 24⅛ reference to “establishments” “suggest[s] industrial sites more fixed and permanent than outdoor timber-harvesting operations.” Ante, at 612. Not so. The *626Standard Industrial Classification uses “establishments” throughout to refer to business entities in general; for example, Classification 2411 refers to “[establishments primarily engaged in cutting timber,” which includes “producing wood chips in the field.” 2 App. 64. I cannot imagine what kind of “fixed and permanent” industrial site the Court and EPA imagine will be “producing wood chips in the field.” And the Court’s final point, ante, at 613—that the regulatory definition of “industrial activity” uses the word “facilities”— cuts the other way: EPA regulations define “facility” to include “any . . . ‘point source.’” 40 CFR §122.2; see, e.g., § 122.26(b)(14)(iii) (referring to mines as “facilities”).
The Agency also assures us that its intent (Brief for United States as Amicus Curiae 25) was to reach a more limited subset of logging activities, an intent that it believes can essentially float free from the text of the relevant rule. In the end, this is the real meat of the matter: EPA states that it simply did not mean to require permits for the discharges at issue here. And the Court is willing to credit that intent, even given what I think has been amply demonstrated to be a contrary text.
* * *
Because the fairest reading of the Agency’s rules proscribes the conduct at issue in these cases, I would affirm the judgment below. It is time for us to presume (to coin a phrase) that an agency says in a rule what it means, and means in a rule what it says there.