delivered the opinion of the Court.
In this case, we decide whether Ring v. Arizona, 536 U. S. 584 (2002), applies retroactively to cases already final on direct review.
*350I
In April 1981, Finance America employee Brenna Bailey disappeared while on a house call to discuss an outstanding debt with respondent Warren Summerlin’s wife. That evening, an anonymous woman (later identified as respondent’s mother-in-law) called the police and accused respondent of murdering Bailey. Bailey’s partially nude body, her skull crushed, was found the next morning in the trunk of her car, wrapped in a bedspread from respondent’s home. Police arrested respondent and later overheard him make incriminating remarks to his wife.
Respondent was convicted of first-degree murder and sexual assault. Arizona’s capital sentencing provisions in effect at the time authorized the death penalty if one of several enumerated aggravating factors was present. See Ariz. Rev. Stat. Ann. §§ 13-703(E), (F) (West 1978), as amended by Act of May 1, 1979 Ariz. Sess. Laws ch. 144. Whether those aggravating factors existed, however, was determined by the trial judge rather than by a jury. § 13-703(B). In this case the judge, after a hearing, found two aggravating factors: a prior felony conviction involving use or threatened use of violence, § 13-703(F)(2), and commission of the offense in an especially heinous, cruel, or depraved manner, § 13-703(F)(6). Finding no mitigating factors, the judge imposed the death sentence. The Arizona Supreme Court affirmed on direct review. State v. Summerlin, 138 Ariz. 426, 675 P. 2d 686 (1983).
Protracted state and federal habeas proceedings followed. While respondent's case was pending in the Ninth Circuit, we decided Apprendi v. New Jersey, 530 U. S. 466 (2000), and Ring v. Arizona, supra. In Apprendi, we interpreted the constitutional due-process and jury-trial guarantees to require that, “[o]ther than the fact of a prior conviction, any fact that increases the penalty for a crime beyond the prescribed statutory maximum must be submitted to a jury, and proved beyond a reasonable doubt.” 530 U. S., at 490. In *351Ring, we applied this principle to a death sentence imposed under the Arizona sentencing scheme at issue here. We concluded that, because Arizona law authorized the death penalty only if an aggravating factor was present, Apprendi required the existence of such a factor to be proved to a jury rather than to a judge. 536 U. S., at 603-609.1 We specifically overruled our earlier decision in Walton v. Arizona, 497 U. S. 639 (1990), which had upheld an Arizona death sentence against a similar challenge. 536 U. S., at 609.
The Ninth Circuit, relying on Ring, invalidated respondent’s death sentence. Summerlin v. Stewart, 341 F. 3d 1082, 1121 (2003) (en banc).2 It rejected the argument that Ring did not apply because respondent’s conviction and sentence had become final on direct review before Ring was decided. We granted certiorari. 540 U. S. 1045 (2003).3
II
When a decision of this Court results in a “new rule,” that rule applies to all criminal cases still pending on direct review. Griffith v. Kentucky, 479 U. S. 314, 328 (1987). As to convictions that are already final, however, the rule applies only in limited circumstances. New substantive rules generally apply retroactively. This includes decisions that narrow the scope of a criminal statute by interpreting its terms, *352see Bousley v. United States, 523 U. S. 614, 620-621 (1998), as well as constitutional determinations that place particular conduct or persons covered by the statute beyond the State’s power to punish, see Saffle v. Parks, 494 U. S. 484, 494-495 (1990); Teague v. Lane, 489 U. S. 288, 311 (1989) (plurality opinion).4 Such rules apply retroactively because they “necessarily carry a significant risk that a defendant stands convicted of ‘an act that the law does not make criminal’” or faces a punishment that the law cannot impose upon him. Bousley, supra, at 620 (quoting Davis v. United States, 417 U. S. 333, 346 (1974)). .
New rules of procedure, on the other hand, generally do not apply retroactively. They do not produce a class of persons convicted of conduct the law does not make criminal, but merely raise the possibility that someone convicted with use of the invalidated procedure might have been acquitted otherwise. Because of this more speculative connection to innocence, we give retroactive effect to only a small set of “ ‘watershed rules of criminal procedure’ implicating the fundamental fairness and accuracy of the criminal proceeding.” Saffle, supra, at 495 (quoting Teague, 489 U. S., at 311 (plurality opinion)). That a new procedural rule is “fundamental” in some abstract sense is not enough; the rule must be one “without which the likelihood of an accurate conviction is seriously diminished.” Id., at 313 (emphasis added). This class of rules is extremely narrow, and “it is unlikely that any ... ‘ha[s] yet to emerge.’” Tyler v. Cain, 533 Ü. S. 656, 667, n. 7 (2001) (quoting Sawyer v. Smith, 497 U. S. 227, 243 (1990)).
The Ninth Circuit agreed with the State that Ring announced a new rule. 341 F. 3d, at 1108-1109. It neverthe*353less applied the rule retroactively to respondent’s case, relying on two alternative theories: first, that it was substantive rather than procedural; and second, that it was a “watershed” procedural rule entitled to retroactive effect. We consider each theory in turn.
A
A ruléis substantive rather than procedural if it alters the range of conduct or the class of persons that the law punishes. See Bousley, supra, at 620-621 (rule “hold[s] that a . . . statute does not reach certain conduct” or “make[s] conduct criminal”); Saffle, supra, at 495 (rule “decriminalize[s] a class of conduct [or] prohibit^] the imposition of . . . punishment on a particular class of persons”). In contrast, rules that regulate only the manner of determining the defendant’s culpability are procedural. See Bousley, supra, at 620.
Judged by this standard, Ring’s holding is properly classified as procedural. Ring held that “a sentencing judge, sitting without a jury, [may not] find an aggravating circumstance necessary for imposition of the death penalty.” 536 U. S., at 609. Rather, “the Sixth Amendment requires that [those circumstances] be found by a jury.” Ibid. This holding did not alter the range of conduct Arizona law subjected to the death penalty. It could not have; it rested entirely on the Sixth Amendment’s jury-trial guarantee, a provision that has nothing to do with the range of conduct a State may criminalize. Instead, Ring altered the range of permissible methods for determining whether a defendant’s conduct is punishable by death, requiring that a jury rather than a judge find the essential facts bearing on punishment. Rules that allocate decisionmaking authority in this fashion are prototypical procedural rules, a conclusion we have reached in numerous other contexts. See Gasperini v. Center for Humanities, Inc., 518 U. S. 415, 426 (1996) (Erie doctrine); Landgraf v. USI Film Products, 511 U. S. 244, 280-281 *354(1994) (antiretroactivity presumption); Dobbert v. Florida, 432 U. S. 282, 293-294 (1977) (Ex Post Facto Clause).
Respondent nevertheless argues that Ring is substantive because it modified the elements of the offense for which he was convicted. He relies on our statement in Ring that, “[b]ecause Arizona’s enumerated aggravating factors operate as ‘the functional equivalent of an element of a greater offense,’ the Sixth Amendment requires that they be found by a jury.” 536 U. S., at 609 (citation omitted); see also Satta-zahn v. Pennsylvania, 537 U. S. 101, 111 (2003) (plurality opinion). The Ninth Circuit agreed, concluding that Ring “reposition[ed] Arizona’s aggravating factors as elements of the separate offense of capital murder and reshaped] the structure of Arizona murder law.” 341 F. 3d, at 1105.
A decision that modifies the elements of an offense is normally substantive rather than procedural. New elements alter the range of conduct the statute punishes, rendering some formerly unlawful conduct lawful or vice versa. See Bousley, 523 U. S., at 620-621. But that is not what Ring did; the range of conduct punished by death in Arizona was the same before Ring as after. Ring held that, because Arizona’s statutory aggravators restricted (as a matter of state law) the class of death-eligible defendants, those aggravators effectively were elements for federal constitutional purposes, and so were subject to the procedural requirements the Constitution attaches to trial of elements. 536 U. S., at 609. This Court’s holding that, because Arizona has made a certain fact essential to the death penalty, that fact must be found by a jury, is not the same as this Court’s making a certain fact essential to the death penalty. The former was a procedural holding; the latter would be substantive. The Ninth Circuit’s conclusion that Ring nonetheless “reshaped] the structure of Arizona murder law,” 341 F. 3d, at 1105, is particularly remarkable in the face of the Arizona Supreme Court’s previous conclusion to the contrary. See State v. *355Towery, 204 Ariz. 386, 390-391, 64 P. 3d 828, 832-833, cert. dism’d, 539 U. S. 986 (2003).5
B
Respondent argues in the alternative that Ring falls under the retroactivity exception for “ ‘watershed rules of criminal procedure’ implicating the fundamental fairness and accuracy of the criminal proceeding.” Saffle, 494 U. S., at 495 (quoting Teague, 489 U. S., at 311). He offers several reasons why juries are more accurate factfinders, including the tendency of group deliberation to suppress individual eccentricities; the jury’s protection from exposure to inadmissible evidence; and its better representation of the common sense of the community. The Ninth Circuit majority added others, including the claim that a judge might be too acclimated to capital sentencing and that he might be swayed by political pressure. 341 F. 3d, at 1109-1116. Respondent further notes that common-law authorities praised the jury’s fact-finding ability. See, e. g., 3 W. Blackstone, Commentaries on the Laws of England 380 (1768); Georgia v. Brailsford, 3 Dall. 1, 4 (1794) (jury charge of Jay, C. J.).
The question here is not, however, whether the Framers believed that juries are more accurate factfinders than judges (perhaps so — they certainly thought juries were more independent, see Blakely v. Washington, ante, at 305-308). Nor is the question whether juries actually are more accurate factfinders than judges (again, perhaps so). Rather, the question is whether judicial factfinding so “seriously dimin-*356ishe[s]” accuracy that there is an ‘“impermissibly large risk’ ” of punishing conduct the law does not reach. Teague, supra, at 312-313 (quoting Desist v. United States, 394 U. S. 244, 262 (1969) (Harlan, J., dissenting)) (emphasis added). The evidence is simply too equivocal to support that conclusion.
First, for every argument why juries are more accurate factfinders, there is another why they are less accurate. The Ninth Circuit dissent noted several, including juries’ tendency to become confused over legal standards and to be influenced by emotion or philosophical predisposition. 341 F. 3d, at 1129-1131 (opinion of Rawlinson, J.) (citing, inter alia, Eisenberg & Wells, Deadly Confusion: Juror Instructions in Capital Cases, 79 Cornell L. Rev. 1 (1993); Garvey, The Emotional Economy of Capital Sentencing, 75 N. Y. U. L. Rev. 26 (2000); and Bowers, Sandys, & Steiner, Foreclosed Impartiality in Capital Sentencing: Jurors’ Predispositions, Guilt-Trial Experience, and Premature Decision Making, 83 Cornell L. Rev. 1476 (1998)). Members of this Court have opined that judicial sentencing may yield more consistent results because of judges’ greater experience. See Proffitt v. Florida, 428 U. S. 242, 252 (1976) (joint opinion of Stewart, Powell, and Stevens, JJ.). Finally, the mixed reception that the right to jury trial has been given in other countries, see Vidmar, The Jury Elsewhere in the World, in World Jury Systems 421-447 (N. Vidmar ed. 2000), though irrelevant to the meaning and continued existence of that right under our Constitution, surely makes it implausible that judicial fact-finding so “seriously diminished]” accuracy as to produce an “ ‘impermissibly large risk’ ” of injustice. When so many presumably reasonable minds continue to disagree over whether juries are better factfinders at all, we cannot confidently say that judicial factfinding seriously diminishes accuracy.
Our decision in DeStefano v. Woods, 392 U. S. 631 (1968) (per curiam), is on point. There we refused to give retroac-*357tiv'e effect to Duncan v. Louisiana, 391 U. S. 145 (1968), which applied the Sixth Amendment’s jury-trial guarantee to the States. While DeStefano was decided under our preTeague retroactivity framework, its reasoning is germane. We noted that, although “the right to jury trial generally tends to prevent arbitrariness and repression!,] . . , ‘[w]e would not assert... that every criminal trial — or any particular trial — held before a judge alone is unfair or that a defendant may never be as fairly treated by a judge as he would be by a jury.’” 392 U. S., at 633-634 (quoting Duncan, supra, at 158). We concluded that “[t]he values implemented by the right to jury trial would not measurably be served by requiring retrial of all persons convicted in the past by procedures not consistent with the Sixth Amendment right to jury trial.” 392 U. S., at 634. If under De-Stefano a trial held entirely without a jury was not impermissibly inaccurate, it is hard to see how a trial in which a judge finds only aggravating factors could be.
The dissent contends that juries are more accurate because they better reflect community standards in deciding whether, for example, a murder was heinous, cruel, or depraved. Post, at 361-362 (opinion of Breyer, J.). But the statute here does not condition death eligibility on whether the offense is heinous, cruel, or depraved as determined by community standards. See Ariz. Rev. Stat. Ann. § 13-703(F)(6) (West 1978). It is easy to find enhanced accuracy in jury determination when one redefines the statute’s substantive scope in such manner as to ensure that result. The dissent also advances several variations on the theme that death is different (or rather, “dramatically different,” post, at 363). Much of this analysis is not an application of Teague, but a rejection of it, in favor of a broader endeavor to “balance competing considerations,” post, at 362. Even were we inclined to revisit Teague in this fashion, we would not agree with the dissent’s conclusions. Finally, the dissent notes that, in DeStefano, we considered factors other than *358enhanced accuracy that are no longer relevant after Teague. See post, at 365. But we held in that case that “fa]H three factors favor only prospective application of the rule.” 392 U. S., at 633 (emphasis added). Thus, the result would have been the same even if enhanced accuracy were the sole criterion for retroactivity.6
* * *
The right to jury trial is fundamental to our system of criminal procedure, and States are bound to enforce the Sixth Amendment’s guarantees as we interpret them. But it does hot follow that, when a criminal defendant has had a full trial and one round of appeals in which the State faithfully applied the Constitution as we understood it at the time, he may nevertheless continue to litigate his claims indefinitely in hopes that we will one day have a change of heart. Ring announced a new procedural rule that does not apply retroactively to cases already final on direct review. The contrary judgment of the Ninth Circuit is reversed, and the case is remanded for further proceedings consistent with this opinion.
It is so ordered.
Because Arizona law already required aggravating factors to be proved beyond a reasonable doubt, see State v. Jordan, 126 Ariz. 283, 286, 614 P. 2d 825, 828, cert. denied, 449 U. S. 986 (1980), that aspect of Apprendi was not at issue.
Because respondent filed his habeas petition before the effective date of the Antiterrorism and Effective Death Penalty Act of 1996, 110 Stat. 1214, the provisions of that Act do not apply. See Lindh v. Murphy, 521 U. S. 320, 336-337 (1997).
The State also sought certiorari on the ground that there was no Apprendi violation because the prior-conviction aggravator, exempt from Apprendi under Almendarez-Torres v. United States, 523 U. S. 224 (1998), was sufficient standing alone to authorize the death penalty. We denied certiorari on that issue, 540 U. S. 1045 (2003), and express no opinion on it.
We have sometimes referred to rules of this latter type as falling under an exception to Teague’s bar on retroactive application of procedural rules, see, e. g., Horn v. Banks, 536 U. S. 266, 271, and n. 5 (2002) (per curiam); they are more accurately characterized as substantive rules not subject to the bar.
Respondent also argues that Ring was substantive because our understanding of Arizona law changed. Compare Ring v. Arizona, 536 U. S. 584, 602-603 (2002), with Apprendi v. New Jersey, 530 U. S. 466, 496-497 (2000). Even if our understanding of state law changed, however, the actual content of state law did not. See State v. Ring, 200 Ariz. 267, 279, 25 P. 3d 1139, 1151 (2001), rev'd on other grounds, 536 U. S. 584 (2002); State v. Gretzler, 135 Ariz. 42, 54, 659 P. 2d 1, 13, cert. denied, 461 U. S. 971 (1983); Johnson v. Fankell, 520 U. S. 911, 916 (1997).
The dissent distinguishes DeStefano on the ground that “this ease involves only a small subclass of defendants deprived of jury trial rights, the relevant harm within that subclass is more widespread, the administration of justice problem is far less serious, and the reliance interest less weighty.” Post, at 366. But the first, third, and fourth of these points are irrelevant under Teague, and the second, insofar as it relates to accuracy, is an unsubstantiated assertion. If jury trial significantly enhances accuracy, we would not have been able to hold as we did in DeStefano that the first factor — “preventing] arbitrariness and repression,” 392 U. S., at 633 — did not favor retroactivity.