dissenting.
Two propositions of law supported our holding in Espinosa v. Florida, 505 U. S. 1079 (1992) (per curiam): First, in a capital sentencing proceeding in a State where the sen-tencer weighs aggravating and mitigating circumstances, the Eighth Amendment is violated by a jury instruction that fails to define the “especially heinous, atrocious, or cruel” (HAC) aggravating circumstance. Second, in a Florida sentencing proceeding the trial court must give “ ‘great weight’ ” to the jury’s recommendation, whether it be for life or death. Id., at 1082. For these reasons, we concluded in Espinosa that constitutional error that taints the jury’s recommendation presumptively taints the judge’s sentence as well. Ibid. The two propositions supporting the Espinosa holding were well established when that case was decided. The first proposition dates back to 1980 when we decided Godfrey v. Georgia, 446 U. S. 420, 428-429,1 and the second was announced by the Florida Supreme Court in *5411975 in Tedder v. State, 322 So. 2d 908, 910.2 Thus I agree with Chief Judge Tjoflat that our per curiam opinion in Es-pinosa amounted to “nothing more than an application of well-settled principles. ... In declaring the Florida HAC instruction unconstitutional, the Court simply applied the law as announced initially in Godfrey and later reaffirmed in Maynard [v. Cartwright, 486 U. S. 356 (1988]). The Court’s conclusion — that the invalid instruction may have tainted the jury’s death penalty recommendation and the trial judge’s sentence — merely acknowledged what the Supreme Court of Florida has been holding for years.” Glock v. Singletary, 65 F. 3d 878, 896 (CA11 1995) (dissenting opinion) (footnotes omitted).3
Today the Court reaches the conclusion that Espinosa announced a new rule by placing a novel interpretation on its holding. The majority apparently construes Espinosa as holding that the constitutional error in the jury instruction will “automatically render a defendant’s sentence unconstitutional.” Ante, at 530.4 The Court suggests that our holdings in Godfrey, Maynard v. Cartwright, 486 U. S. 356 (1988), and Clemons v. Mississippi, 494 U. S. 738, 745 (1990) — that *542an appellate court could cure a sentencing jury’s weighing of an invalid aggravator — might have led a reasonable jurist down a road different from the one the Court followed in Espinosa.5 But in holding that a trial judge’s sentence may be infected by the jury’s consideration of an invalid aggravating factor, Espinosa did not address the entirely separate question of whether the jury’s error could be cured or considered harmless either at the trial or the appellate level. Indeed, in subsequent proceedings the Supreme Court of Florida did conclude that the error in Espinosa’s case was harmless and upheld his sentence of death. See Espinosa v. State, 626 So. 2d 165, 167 (1993) (ruling that Espinosa’s HAC instruction claim was procedurally barred because he had challenged the HAC factor rather than the instruction itself and, alternatively, that any error in the instruction was harmless beyond a reasonable doubt), cert. denied, 511 U. S. 1152 (1994), and affirmed Espinosa’s sentence. Our decision in Espinosa did not create a new rule prohibiting trial courts from curing a jury’s error, rather it held that “if a weighing State decides to place capital sentencing authority in two actors rather than one, neither actor must be permitted to weigh invalid aggravating circumstances.” 505 U. S., at 1082. This holding is a logical consequence of applying Godfrey to Florida’s sentencing scheme.
In a sinuous, difficult to follow argument, the Court suggests that three hypothetical propositions of law somehow demonstrate that the narrow holding in Espinosa was not dictated by Godfrey and Tedder. First, the Court posits that a reasonable jurist might have believed that “[t]he mere cabining of the trial court’s discretion” was alone enough to avoid constitutional error. Ante, at 532 (emphasis deleted). *543A critical part of that “cabining,” however, is Florida’s requirement that a properly instructed jury must have an opportunity to recommend a life sentence, and that the judge must give great weight to that recommendation. The role of the jury is to provide one of the cabin’s four walls. The fact that three walls remain standing hardly excuses an error that removed the wall represented by the jury’s recommendation. At the time of petitioner’s sentencing, the Florida Supreme Court recognized the jury’s critical role, and, when error occurred before the jury, did not hesitate to remand for resentencing, even when the trial judge claimed to be unaffected by the error. For example in Messer v. State, 330 So. 2d 137, 142 (1976), the State Supreme Court remanded for resentencing when the trial court failed to allow the jury to consider certain mitigating evidence. The court rejected the argument that the trial court’s subsequent weighing of the mitigating evidence cured the error: The Florida scheme, the court concluded, was one of “checks and balances in which the input of the jury serves as an integral part.” Ibid. Our holding in Proffitt v. Florida, 428 U. S. 242, 255 (1976) (joint opinion), that Florida’s sentencing scheme is not facially unconstitutional does not suggest otherwise. There, we determined that the State’s sentencing procedure provided adequate safeguards against arbitrary imposition of the death sentence in part because of the procedures followed by the trial judge in fixing the sentence. Our focus was on the adequacy of the guidance provided by the sentencing scheme; accordingly, we had no need to extensively examine or discuss the judge’s relationship to the jury or Florida Supreme Court decisions like Tedder.
Second, simply ignoring the reasoning in Tedder, the Court suggests that there was “no error for the trial judge to cure, since under Florida law the trial court, not the jury, was the sentencer.” Ante, at 533 (emphasis deleted). It is, of course, true that the judge imposes the sentence after receiving the jury’s recommendation. But this has never *544meant that constitutional error in the proceedings before the jury is simply irrelevant. Cf. Messer v. State, supra. As then-JusTiCE Rehnquist noted in 1983, it is well-settled Florida law that if the jury makes a recommendation of life imprisonment, “the trial judge may not impose a death sentence unless ‘the facts suggesting a sentence of death [are] so clear and convincing that virtually no reasonable person could differ.’ Tedder v. State, 322 So. 2d 908, 910 (1975).” Barclay v. Florida, 463 U. S. 939, 955-956 (plurality opinion).6 Similarly, a trial judge should not disturb a jury recommendation of death “unless there appear strong reasons to believe that reasonable persons could not agree with the recommendation.” See LeDuc v. State, 365 So. 2d 149, 151 (Fla. 1978), cert. denied, 444 U. S. 885 (1979). Given this, it is vacuous to argue that our prior references to the judge as the sentencer somehow imply that an error before the jury would not affect the ultimate sentence. It is equally vacuous to suggest that our conclusion in Espinosa “that the jury was at least in part a cosentencer” had its source in a case decided “just three weeks earlier,” ante, at 533 (citing Sochor v. Florida, 504 U. S. 527 (1992)). In that earlier case, we cited Tedder after explaining that the jury was a constituent element of the sentencer “because the trial judge does not render wholly independent judgment, but must accord deference to the jury’s recommendation.” Sochor, 504 U. S., at 533.
Third, the Court suggests that the trial court’s “weighing of properly narrowed aggravators and mitigators was sufficiently independent of the jury to cure any error in the jury’s consideration of a vague aggravator.” Ante, at 534 (em*545phasis deleted). This suggestion is doubly flawed. Given that the judge’s instruction to the jury failed to narrow the HAC aggravator, there is no reason to believe that he appropriately narrowed the factor in his own deliberations.7 More importantly, even if he did apply a limiting definition, his sentencing decision was made without the benefit of an untainted recommendation from the jury, and, under Florida law, he could not have simply resentenced the petitioner without -regard to the jury’s tainted recommendation. Nor can one simply conclude that this error made no practical difference in petitioner’s sentence. There is nothing in the record to suggest that had the jury recommended a life sentence, the judge would have found that “the facts suggesting a sentence of death were so clear and convincing that virtually no reasonable person could differ,” as Tedder requires.
Here, again, the Court finds that our statements in cases like Walton v. Arizona, 497 U. S. 639 (1990), that a state appellate court may affirm a death sentence resulting from an unconstitutionally broad aggravator by applying a limiting definition, suggest that Espinosa is a new rule. The majority’s analysis confuses an appellate court’s application of a limiting definition on appellate review with a trial judge’s deference to a tainted jury recommendation. The judge in this case did not indicate that he was applying a limiting definition of the HAC factor, or that he was in some other way curing or discounting the error in the jury instruction. At the time of petitioner’s sentencing, given Godfrey and Tedder, this rendered petitioner’s death sentence constitutionally defective.
As a matter of logic and law there was nothing new about Espinosa’s holding that the jury plays a central role in Florida’s capital sentencing scheme. Moreover, as statistics that *546I have previously summarized demonstrate, it was equally clear as a matter of fact that “erroneous instructions to the jury at the sentencing phase of the trial may make the difference between life or death.” Sochor v. Florida, 504 U. S., at 552.8
I respectfully dissent.
Godfrey, of course, held that Georgia’s “outrageously or wantonly vile, horrible and inhuman” aggravating factor failed to adequately channel the jury’s discretion. See 446 U. S., at 428-429. We found the “heinous, atrocious or cruel” aggravator unconstitutional in Maynard v. Cartwright, 486 U. S. 356, 359 (1988), and subsequently noted that application of Godfrey to the HAC instruction did not create a new rule. See Stringer v. Black, 503 U. S. 222, 228-229 (1992).
These two “controlling precedents,” both of which were cited in the Espinosa opinion, provided sufficient support for its holding. Thus the Court is simply mistaken when it asserts that “Espinosa itself did not purport to rely upon any controlling precedent.” Ante, at 528.
Tedder, of course, was not an isolated decision. In Riley v. Wainwright, 517 So. 2d 656 (Fla. 1987), the State Supreme Court put the point succinctly: “If the jury’s recommendation, upon which the judge must rely, results from an unconstitutional procedure, then the entire sentencing process necessarily is tainted by that procedure.” Id., at 659. The Riley court relied on a pre-Tedder decision stating that the advisory opinion of the jury “is an integral part of the death sentencing process.” 517 So. 2d, at 657 (citing Lamadline v. State, 303 So. 2d 17, 20 (Fla. 1974)).
Responding to this dissent in n. 2, ante, at 528, the Court states that the clause I have quoted was not intended to describe the Court’s understanding of the holding in Espinosa. If that be so, the relevance of this portion of the Court’s opinion, including its reliance on Godfrey and Maynard, is opaque, at best.
The Court also relies heavily on a passage in our opinion in Walton v. Arizona, 497 U. S. 639 (1990), noting that a trial judge’s failure to apply a narrowing construction to an invalid aggravator ‘“does not necessarily require that a state appellate court vacate a death sentence based on that factor.’” Ante, at 537.
The Florida Supreme Court has applied Tedder in numerous cases to reverse a trial judge’s override of a jury’s life sentence. See, e. g., Wasko v. State, 505 So. 2d 1314, 1318 (1987); Goodwin v. State, 405 So. 2d 170, 172 (1981); Odom v. State, 403 So. 2d 936, 942-943 (1981), cert. denied, 456 U. S. 925 (1982); Neary v. State, 384 So. 2d 881, 885-886 (1980); Malloy v. State, 382 So. 2d 1190, 1193 (1979).
Nothing in the record indicates that the judge recognized that the jury instruction was erroneous, or that he sought to cure that error in his own weighing process. In finding that the HAC aggravator was present, the judge merely stated: “The facts speak for themselves.” App. 20.
“As a matter of feet, the jury sentence is the sentence that is usually imposed by the Florida Supreme Court. The State has attached an appendix to its brief, see App. to Brief for Respondent A1-A70, setting forth data concerning 469 capital eases that were reviewed by the Florida Supreme Court between 1980 and 1991. In 341 of those eases (73%), the jury recommended the death penalty; in none of those cases did the trial judge impose a lesser sentence. In 91 cases (19%), the jury recommended a life sentence; in all but one of those cases, the trial judge overrode the jury’s recommended life sentence and imposed a death sentence. In 69 of those overrides (77%), however, the Florida Supreme Court vacated the trial judge’s sentence and either imposed a life sentence itself or remanded for a new sentencing hearing.
“Two conclusions are evident. First, when the jury recommends a death sentence, the trial judge will almost certainly impose that sentence. Second, when the jury recommends a life sentence, although overrides have been sustained occasionally, the Florida Supreme Court will normally uphold the jury rather than the judge. It is therefore clear that in practice, erroneous instructions to the jury at the sentencing phase of the trial may make the difference between life or death.” Sochor v. Florida, 504 U. S., at 551-552 (footnote omitted).