delivered the opinion of the Court.
In Mills v. Maryland, 486 U. S. 867 (1988), and McKoy v. North Carolina, 494 U. S. 433 (1990), this Court held invalid capital sentencing schemes that require juries to disregard mitigating factors not found unanimously. In this case, we must determine whether the rule announced , in Mills and McKoy can be applied on federal habeas corpus review to a defendant whose conviction became final in 1987. Under our retroactivity analysis as set forth in Teague v. Lane, 489 U. S. 288 (1989), federal habeas corpus petitioners may not avail themselves of new rules of criminal procedure outside two narrow exceptions. We conclude that Mills announced a new rule that does not. fall within either of Teague’s exceptions.
I
More than 20 years ago, a jury convicted respondent, George Banks, of 12 counts of first-degree murder, and the trial court sentenced him to death. The facts of this case are set forth in detail in the Pennsylvania Supreme Court’s decision affirming respondent’s conviction and sentence on direct review. See Commonwealth v. Banks, 513 Pa. 318, 521 A. 2d 1 (1987). Direct review ended when this Court denied certiorari on October 5, 1987. Banks v. Pennsylvania, 484 U. S. 873. Approximately eight months later, this Court handed down its decision in Mills, supra, which announced that the Constitution forbids States from imposing *409a requirement that the jury find a potential mitigating factor unanimously before that factor may be considered in the sentencing decision.
Respondent pursued state postconviction relief on the theory that the instructions and verdict form given to the jury in his case violated the Mills principle, but the Pennsylvania Supreme Court rejected this claim on the merits. See Commonwealth v. Banks, 540 Pa. 143, 656 A. 2d 467 (1995). Respondent then turned to the federal courts. Although the District Court denied relief, Banks v. Horn, 63 F. Supp. 2d 525 (MD Pa. 1999), the Court of Appeals for the Third Circuit reversed respondent’s death sentence, Banks v. Horn, 271 F. 3d 527 (2001). In reaching its decision, the Court of Appeals declined to apply the retroactivity analysis set forth in Teague v. Lane, supra, to the question whether Mills applied retroactively to respondent. This was not necessary, in the Court of Appeals’ view, because the Pennsylvania Supreme Court had itself applied Mills. 271 F. 3d, at 543. We summarily reversed, holding that “in addition to performing any analysis required by AEDPA [the Antiterrorism and Effective Death Penalty Act of 1996], a federal court considering a habeas petition must conduct a threshold Teague analysis when the issue is properly raised by the state.” Horn v. Banks, 536 U. S. 266, 272 (2002) (per curiam) (Banks I):
On remand, the Court of Appeals considered the retroactive application of Mills. Banks v. Horn, 316 F. 3d 228 (CA3 2003). The court recognized that its primary task was to determine whether Mills announced a new rule, and that this, in turn, required it to ascertain whether the precedent existing at the time respondent’s conviction became final dictated or compelled the rule in Mills. 316 F. 3d, at 233-235. From this Court’s decisions in Lockett v. Ohio, 438 U. S. 586 (1978), Eddings v. Oklahoma, 455 U. S. 104 (1982), and their direct progeny, the Court of Appeals distilled the rule that the “Eighth Amendment prohibits any barrier to the sen-tencer’s consideration of mitigating evidence.” 316 F. 3d, at *410289. The Court of Appeals characterized this Court’s decision in Mills as “merely recognizing] that the perceived need for unanimity could constitute one such unconstitutional barrier,” and concluded that the existing legal landscape compelled the decision in Mills. 316 F. 3d, at 240. Accordingly, the court held that Mills applied retroactively to respondent and reinstated the remainder of its previous opinion, again granting respondent relief from his death sentence.1
We granted the Commonwealth’s second petition for certiorari in this case to decide whether Mills applies retroactively to respondent and, if so, whether the Pennsylvania Supreme Court unreasonably applied federal law in holding that there was no Mills error in respondent’s case. 539 U. S. 987 (2003). Although the Lockett/Eddings line of cases supports the Court’s decision in Mills, it does not compel that decision. Mills therefore announced a new rule. We are also unable to conclude that the Mills rule falls under either Teague exception. In particular, Mills did not announce a “watershed rul[e] of criminal procedure implicating the fundamental fairness and accuracy of the criminal proceeding.” Saffle v. Parks, 494 U. S. 484, 495 (1990) (internal quotation marks omitted). Accordingly, we again reverse the judgment of the Court of Appeals.2
*411II
Under Teague, the determination whether a constitutional rule of criminal procedure applies to a case on collateral review involves a three-step process. See, e. g., Lambrix v. Singletary, 520 U. S. 518, 527 (1997). First, the court must determine when the defendant’s conviction became final. Second, it must ascertain the “legal landscape as it then existed,” Graham v. Collins, 506 U. S. 461, 468 (1993), and ask whether the Constitution, as interpreted by the precedent then existing, compels the rule, Saffle, supra, at 488. That is, the court must decide whether the rule is actually “new.” Finally, if the rule is new, the court must consider whether it falls within either of the two exceptions to nonretroactivity. Lambrix, supra, at 527.3
A
Ordinarily, ascertaining the date on which a defendant’s conviction becomes final poses no difficulties: State convictions are final “for purposes of retroactivity analysis when the availability of direct appeal to the state courts has been exhausted and the time for filing a petition for a writ of certiorari has elapsed or a timely filed petition has been finally denied.” Caspari v. Bohlen, 510 U. S. 383, 390 (1994). See also Clay v. United States, 537 U. S. 522, 527 (2003). Respondent, however, urges a different rule. He argues that in view of the Pennsylvania Supreme Court’s unique “relaxed waiver rule” — pursuant to which that court considered his Mills claim on the merits — his conviction became final for Teague purposes in 1995 when the State Supreme Court decided the Mills claim against him. Brief for Respondent 25-31. Because of the Pennsylvania Supreme Court’s practice *412of considering forfeited claims in capital cases, respondent insists, “conventional notions of ‘finality’” do not apply. Id., at 27.
In the past, the Pennsylvania Supreme Court did, in fact, apply a “relaxed waiver rule” in death penalty cases. See, e. g., Commonwealth v. DeHart, 539 Pa. 5, 25, 650 A. 2d 38, 48 (1994); Commonwealth v. Billa, 521 Pa. 168, 181, 555 A. 2d 835, 842 (1989). But this practice, which the court has abandoned, see Commonwealth v. Albrecht, 554 Pa. 31, 44-46, 720 A. 2d 693, 700 (1998), “was not absolute, but discretionary,” Commonwealth v. Freeman, 573 Pa. 532, 557, n. 9, 827 A. 2d 385, 400, n. 9 (2003) (describing past practice). Notably, the Pennsylvania Supreme Court has expressly stated, in a capital case, that it would decline to apply Mills retroactively. Commonwealth v. Peterkin, 538 Pa. 455, 465, n. 4, 649 A. 2d 121, 126, n. 4 (1994).
A state court’s past discretionary “‘practice’ [of] declining] to apply ordinary waiver principles in capital cases,” Albrecht, supra, at 44, 720 A. 2d, at 700, does not render convictions and sentences that are no longer subject to direct review nonfinal for Teague purposes. Such a judgment is “final” despite the possibility that a state court might, in its discretion, decline to enforce an available procedural bar and choose to apply a new rule of law. Cf. Wainwright v. Sykes, 433 U. S. 72, 81-91 (1977).
Respondent’s argument reflects a fundamental misunderstanding of Teague. Teague's nonretroactivity principle acts as a limitation on the power of federal courts to grant “ha-beas corpus relief to . . . state prisoner[s].” Caspari, 510 U. S., at 389. That is why federal habeas corpus courts “must apply Teague before considering the merits of [a] claim,” ibid., whenever the State raises the question, a point we explained in Banks I, see 536 U. S., at 271. See also id., at 271-272 (explaining that the Court of Appeals had erred by focusing only on the Pennsylvania Supreme Court’s treatment of respondent’s Mills claim).
*413This should make clear that the Teague principle protects not only the reasonable judgments of state courts but also the States’ interest in finality quite apart from their courts. As Teague explained:
“In many ways the application of new rules to eases on collateral review may be more intrusive than the enjoining of criminal prosecutions, cf. Younger v. Harris, 401 U. S. 37, 43-54 (1971), for it continually forces the States to marshal resources in order to keep in prison defendants whose trials and appeals conformed to then-existing constitutional standards.” 489 U. S., at 310.
In short, our rule for determining when a state conviction becomes final applies to this case without modification, and we agree with the Court of Appeals that respondent’s conviction became final in 1987. See 316 F. 3d, at 235.
B
We must therefore assay the legal landscape as of 1987 and ask “whether the rule later announced in [Mills] was dictated by then-existing precedent — whether, that is, the unlawfulness of [respondent’s] conviction was apparent to all reasonable jurists.” Lambrix, supra, at 527-528. In Mills, the Court held that the Constitution prohibits States from requiring jurors to find mitigating factors unanimously. McKoy, 494 U. S., at 444; Mills, 486 U. S., at 374-375; id., at 384 (vacating death sentence because the jury instructions gave rise to a “substantial probability that reasonable jurors ... may have thought they were precluded from considering any mitigating evidence” not found unanimously).4
In reaching its conclusion, the Court in Mills and McKoy relied on a line of cases beginning with Lockett v. Ohio, 438 *414U. S. 586 (1978) (plurality opinion), and Eddings v. Oklahoma, 455 U. S. 104 (1982). In Lockett, a plurality of the Court struck down Ohio’s death penalty statute because it prevented the sentencer from “considering, as a mitigating factor,”, certain “aspect[s] of a defendant’s character or record and [certain] circumstances of the offense that the defendant profferfed] as a basis for a sentence less than.death.” 438 U. S., at 604. A majority of the Court first embraced this principle in Eddings. There, the Court confronted a situation in which the sentencer had found, “as a matter of law [that it] was unable even to consider [potentially mitigating] evidence.” 455 U. S., at 113. The Court held that this limitation violated the Lockett rule. 455 U. S., at 113-115. See also Skipper v. South Carolina, 476 U. S. 1, 4, 8-9 (1986) (holding that States cannot, through evidentiary rules, exclude relevant mitigating evidence from the sentencer’s consideration).
In Mills, the Court noted that its previous eases did not depend on the source of the potential barrier to the sen-tencer’s ability to consider mitigating evidence. 486 U. S., at 375. The Court then asserted that “[t]he same [rule must apply] with respect to a single juror’s holdout vote against finding the presence of a mitigating circumstance.” Ibid. See also McKoy, supra, at 441-443 (quoting Mills and performing the same analysis).
The generalized Lockett rule (that the sentencer must be allowed to consider any mitigating evidence) could be thought to support the Court’s conclusion in Mills and McKoy. But what is essential here is that it does not mandate the Mills rule. Each of the eases relied on by Mills (and McKoy) specifically considered only obstructions to the sentencer’s ability to consider mitigating evidence. Mills’ innovation rests with its shift in focus to individual jurors. We think it clear that reasonable jurists could have differed as to whether the Lockett principle compelled Mills. See Lambrix, 520 U S., at 527-528.
*415But there is no need to guess. In Mills, four Justices dissented, reasoning that because nothing prevented the jurors from hearing any mitigating evidence that the defendant proffered, the Lockett principle did not control. 486 U. S., at 394 (opinion of Rehnquist, C. J.). In McKoy, three Justices dissented, explaining that “‘the principle established in Lockett does not remotely support” the new focus on individual jurors. 494 U. S., at 464 (opinion of Scalia, J.); see id., at 466 (“In short, Lockett and Eddings are quite simply irrelevant to the question before us . . .”); see also id., at 452-453 (Kennedy, J., concurring in judgment) (noting that the Court “stretche[d]” the Lockett cases “beyond their proper bounds”). The dissent in McKoy stressed the Court’s move from jury to juror. See 494 U. S., at 465-466 (opinion of Scalia, J.). Indeed, prior to Mills, none of the Court’s relevant cases addressed individual jurors, see, e. g., Hitchcock v. Dugger, 481 U. S. 393 (1987), a trend that continued even after Mills, see, e. g., Saffle v. Parks, 494 U. S. 484 (1990); Penry v. Lynaugh, 492 U. S. 302 (1989); Franklin v. Lynaugh, 487 U. S. 164 (1988).
The McKoy dissent also explained that the Mills rule governs how the sentencer considers evidence, not what evidence it considers. In the dissent’s view, the Lockett line governed the latter but not the former. See 494 U. S., at 465-466 (opinion of Scalia, J.). For this distinction, the dissent relied on Saffle v. Parks, supra, decided the same day. There, the Court held that the Lockett line of cases did not compel (assuming it informed) the sought-for rule that States may not “instruct the sentencer to render its decision on the evidence without sympathy.” 494 U. S., at 490. The Court observed:
“Parks asks us to create a rule relating, not to what mitigating evidence the jury must be permitted to consider in making its sentencing decision, but to how it must consider the mitigating evidence. There is a simple and logical difference between rules that govern *416what factors the jury must be permitted to consider in making its sentencing decision and rules that govern how the State may guide the jury in considering and weighing those factors in reaching a decision.” Ibid.
Thus, although the Lockett principle — conceived of at a high level of generality — could be thought to support the Mills rule, reasonable jurists differed even as to this point. It follows a fortiori that reasonable jurists could have concluded that the Lockett line of cases did not compel Mills5 Given the brand new attention Mills paid to individual jurors and the relevance of the what/how distinction drawn in Saffle (which again distinguishes Mills from the Lockett line), we must conclude that the Mills rule “br[o]k[e] new ground,” Teague, 489 U. S., at 301.6 Accordingly, Mills announced a new rule, which does not apply to respondent on collateral review, unless, of course, it falls under one of Teague's exceptions.
C
Teague's bar on retroactive application of new rules of constitutional criminal procedure has two exceptions. First, the bar does not apply to rules forbidding punishment “of certain primary conduct [or to] rules prohibiting a certain category of punishment for a class of defendants because of their status or offense.” Penry, supra, at 330; see also *417O’Dell v. Netherland, 521 U. S. 151, 157 (1997).7 There is no argument that this exception applies here. The second exception is for “ ‘watershed rules of criminal procedure implicating the fundamental fairness and accuracy of the criminal proceeding/ ” Ibid, (quoting Graham, 506 U. S., at 478).
We have repeatedly emphasized the limited scope of the second Teague exception, explaining that “‘it is clearly meant to apply only to a small core of rules requiring observance of those procedures that... are implicit in the concept of ordered liberty/ ” O’Dell, supra, at 157 (quoting Graham, supra, at 478). And, because any qualifying rule “‘would be so central to an accurate determination of innocence or guilt [that it is] unlikely that many such components of basic due process have yet to emerge/” Graham, supra, at 478 (quoting Teague, supra, at 313), it should come as no surprise that we have yet to find a new rule that falls under the second Teague exception. Perhaps for this reason, respondent does not even attempt to argue that Mills qualifies or to rebut petitioners’ argument that it does not, Brief for Petitioners 23-26.
In providing guidance as to what might fall within this exception, we have repeatedly referred to the rule of Gideon v. Wainwright, 372 U. S. 335 (1963) (right to counsel), and only to this rule. See, e. g., Saffle, supra, at 495; cf. Gilmore v. Taylor, 508 U. S. 333, 364 (1993) (Blackmun, J., dissenting). Gideon overruled Betts v. Brady, 316 U. S. 455 (1942), noting that Betts itself had “made an abrupt break with [the Court’s] well-considered precedents.” 372 U. S., at 344. The Court continued:
“Lawyers to prosecute are everywhere deemed essential to protect the public’s interest in an orderly society. Similarly, there are few defendants charged with crime, few indeed, who fail to hire the best lawyers they can *418get to prepare and present their defenses. That government hires lawyers to prosecute and defendants who have the money hire lawyers to defend are the strongest indications of the widespread belief that lawyers in criminal courts are necessities, not luxuries. The right of one charged with crime to counsel may not be deemed fundamental and essential to fair trials in some countries, but it is in ours. From the very beginning, our state and national constitutions and laws have laid great emphasis on procedural and substantive safeguards designed to assure fair trials before impartial tribunals in which every defendant stands equal before the law. This noble ideal cannot be realized if the poor man charged with crime has to face his accusers without a lawyer to assist him.” Ibid, (emphasis added).
See also id., at 344-345 (quoting Powell v. Alabama, 287 U. S. 45, 68-69 (1932)). Gideon, it is fair to say, “alter[ed] our understanding of the bedrock procedural elements essential to the fairness of a proceeding.” Sawyer v. Smith, 497 U. S. 227, 242 (1990) (internal quotation marks omitted).
By contrast, we have not hesitated to hold that less sweeping and fundamental rules do not fall within Teague’s second exception. In O’Dell v. Netherland, supra, for example, we considered the retroactivity of the rule announced in Simmons v. South Carolina, 512 U. S. 154 (1994). Simmons held that a capital defendant must be allowed to inform the sentencer that he would be ineligible for parole if the prosecution argues future dangerousnéss. We rejected the petitioner’s argument that the Simmons rule was “‘on par’ with Gideon v. Wainwright, 372 U. S. 335 (1963),” emphasizing “the sweeping [nature] of Gideon, which established an affirmative right to counsel in all felony cases.” O’Dell, supra, at 167.
And, in Sawyer v. Smith, supra, we considered whether a habeas petitioner could make use of the rule announced in Caldwell v. Mississippi, 472 U. S. 320, 323 (1985) (holding *419that the Eighth Amendment bars imposition of the death penalty by a jury that had been led to believe that responsibility for the ultimate decision rested elsewhere). There too we declined to give retroactive effect to a rule that effectively withheld relevant information from the sentences See Sawyer, supra, at 242-245. We acknowledged that the Caldwell rule was intended to enhance “the accuracy of capital sentencing.” 497 U. S., at 244. But because it effected an incremental change, we could not conclude that “this systemic rule enhancing reliability is an ‘absolute prerequisite to fundamental fairness.’ ” Ibid, (quoting Teague, 489 U. S., at 314). See also Graham, supra, at 478 (concluding that the rule announced in Penry v. Lynaugh, 492 U. S. 302 (1989), does not fall within the second Teague exception).
We recognize that avoidance of potentially arbitrary impositions of the death sentence motivated the Court in Mills and McKoy. Mills described two troubling situations that could theoretically occur absent the Mills rule. Eleven of twelve jurors, could, for example, agree that six mitigating circumstances existed, but one holdout juror could nevertheless force the death sentence. Similarly, all 12 jurors could agree that some mitigating circumstances existed and that these outweighed any aggravators, but because they did not agree on which mitigating circumstances were present, they would again have to return a death sentence. See Mills, 486 U. S., at 373-374; see also McKoy, 494 U. S., at 439-440 (describing these examples). Imposition of the death penalty in these circumstances, the Court reasoned, “would be the ‘height of arbitrariness.’” Id., at 440 (quoting Mills, supra, at 374). See also McKoy, supra, at 454 (Kennedy, J., concurring in judgment).
Quite obviously, the Court decided Mills and McKoy as it did to avoid this possibility. But because “[a]ll of our Eighth Amendment jurisprudence concerning capital sentencing is directed toward the enhancement of reliability and accuracy in some sense,” the fact that a new rule removes some re*420mote possibility of arbitrary infliction of the death sentence does not suffice to bring it within Teague’s second exception. Sawyer, supra, at 243.
However laudable the Mills rule might be, “it has none of the primacy and centrality of the rule adopted in Gideon.” Saffle, 494 U. S., at 495. The Mills rule applies fairly narrowly and works no fundamental shift in “our understanding of the bedrock procedural elements” essential to fundamental fairness. O’Dell, 521 U. S., at 167 (internal quotation marks omitted). We therefore conclude that the Mills rule does not fall within the second Teague exception.
m
We hold that Mills announced a new rule of constitutional criminal procedure that falls within neither Teague exception. Accordingly, that rule cannot be applied retroactively to respondent. The judgment of the Court of Appeals is reversed, and the case is remanded for further proceedings consistent with this opinion.
It is so ordered.
Judge Sloviter wrote separately to express her view that Mills v. Maryland, 486 U. S. 367 (1988), established a new rule that qualified for neither Teague v. Lane, 489 U. S. 288 (1989), exception. 316 F. 3d 228, 253-254 (CA3 2003) (opinion concurring in judgment). Judge Sloviter nevertheless posited that Mills could be applied to respondent because of Pennsylvania’s “unique relaxed waiver doctrine in capital cases.” 316 F. 3d, at 256.
Given our determination that the Court of Appeals erred in holding that Mills applied retroactively to respondent, we do not reach the question whether the Court of Appeals also erred in concluding that the Pennsylvania Supreme Court unreasonably applied Mills.
Rules that fall within what we have referred to as Teague’s first exception “are more accurately characterized as substantive rules not subject to [Teague’s] bar.” Schriro v. Summerlin, ante, at 352, n. 4. See also infra, at 416, and n. 7.
Although nothing in this case turns on. it, we note that it is arguable that the “Mills rule” did not fully emerge until the Court issued McKoy v. North Carolina, 494 U. S. 433 (1990). See id., at 459-463 (Scalia, J., dissenting).
Because the focus of the inquiry is whether reasonable jurists could differ as to whether precedent compels the sought-for rule, we do not suggest that the mere existence of a dissent suffices to show that the rule is new.
The Court of Appeals erred by drawing from Lockett v. Ohio, 438 U. S. 586 (1978), and Eddings v. Oklahoma, 455 U. S. 104 (1982), the general rule that “the Constitution prohibited any barrier to the jury’s consideration of mitigating evidence,” 316 F. 3d, at 241-243 (emphasis added), without also acknowledging that the rule, for purposes of the Teague analysis, did not automatically extend to arguably analogous contexts. It is with respect to this last point that reasonable jurists did in fact differ.
As noted above, these rules are more properly viewed as substantive and therefore not subject to Teague’s bar. See n. 3, supra.