with whom Justice Marshall joins, and with whom Justice Blackmun and Justice Stevens join as to Parts I through IV-A, dissenting.
The murders that Gary Tison and Randy Greenawalt committed revolt and grieve all who learn of them. When the deaths of the Lyons family and Theresa Tyson were first reported, many in Arizona erupted “in a towering yell” for retribution and justice.1 Yet Gary Tison, the central figure in this tragedy, the man who had his family arrange his and Greenawalt’s escape from prison, and the man who chose, with Greenawalt, to murder this family while his sons stood by, died of exposure in the desert before society could arrest him and bring him to trial. The question this case presents is what punishment Arizona may constitutionally exact from two of Gary Tison’s sons for their role in these events. Because our precedents and our Constitution compel a different answer than the one the Court reaches today, I dissent.
HH
Under the felony-murder doctrine, a person who commits a felony is liable for any murder that occurs during the commission of that felony, regardless of whether he or she commits, attempts to commit, or intended to commit that murder. The doctrine thus imposes liability on felons for killings committed by cofelons during a felony. This curious doctrine is a living fossil from a legal era in which all felonies were punishable by death; in those circumstances, the state of mind of the felon with respect to the murder was understandably superfluous, because he or she could be executed simply for intentionally committing the felony.2 Today, in *160most American jurisdictions and in virtually all European and Commonwealth countries, a felon cannot be executed for a murder that he or she did not commit or specifically intend or attempt to commit. In some American jurisdictions, however, the authority to impose death in such circumstances still persists. Arizona is such a jurisdiction.
The proceedings below illustrate how, under the felony-murder doctrine, a defendant may be held liable and sentenced to death for a murder that he or she neither committed nor intended to commit. The prosecutor argued to the jury that it did not matter that Gary Tison and Randy Greenawalt had caused the killings, because under the felony-murder rule the Tisons could nonetheless be found legally responsible for those killings. App. 173-174, 185, 191. The trial judge’s instructions were consistent with the prosecutor’s argument. Id., at 179, 218-219. In sentencing petitioners, the trial court did not find that they had killed, attempted to kill, or intended to kill anyone. Id., at 280-289. Nevertheless, the court upheld the jury’s verdict that Ricky and Raymond Tison were liable under the felony-murder doctrine for the murders that their father and Randy Green-awalt had committed. Furthermore, the court found as an aggravating factor against petitioners the “heinous, cruel and depraved manner” in which Gary Tison and Randy Green-awalt carried out the murders. Id., at 282-283. As a result, the court imposed the death sentence.3
*161The Arizona Supreme Court affirmed. It held that the Tisons “did not specifically intend that the Lyons and Theresa Tyson die, that they did not plot in advance that these homicides would take place, [and] that they did not actually pull the triggers on the guns which inflicted the fatal wounds. ...” State v. Tison, 129 Ariz. 526, 545, 633 P. 2d 335, 354 (1981). The court found these facts to be “of little significance,” however, because “the non-participation in the shooting was not controlling since both [brothers] took part in the robbery, the kidnapping, and were present assisting in the detention of the Lyonses and Theresa Tyson while the homicides were committed.” State v. Tison, 129 Ariz. 546, 556, 633 P. 2d 355, 365 (1981). Thus, while the Arizona courts acknowledged that petitioners had neither participated in the shootings nor intended that they occur, those courts nonetheless imposed the death sentence under the theory of felony murder.
After the decision of the Arizona Supreme Court, this Court addressed, in Enmund v. Florida, 458 U. S. 782 (1982), the question “whether death is a valid penalty under the Eighth and Fourteenth Amendments for one who neither took life, attempted to take life, nor intended to take life.” Id., at 787. The question arose because the Florida Supreme Court affirmed the death sentence for Earl En-mund, an accomplice in an armed robbery in which his two cofelons had killed the two individuals that the felons had intended to rob. Enmund did not shoot anyone, and there was nothing in the record concerning Enmund’s mental state with regard to the killings, but the Florida Supreme Court had held him strictly hable for the killings under the felony-murder doctrine. Enmund v. State, 399 So. 2d 1362, 1369 (1981).
*162In reversing the Florida Supreme Court, this Court took note of the “overwhelming evidence” of “[society’s rejection of the death penalty for accomplice liability in felony murders.” 458 U. S., at 794. The Court observed that, in imposing the death penalty upon Enmund, the Florida Supreme Court had failed to focus on “Enmund’s own conduct . . . [and] on his culpability.” Id., at 798 (emphasis in original). The Court then explained, and rejected, the felony-murder doctrine as a theory of capital culpability.
“Enmund did not kill or intend to kill and thus his culpability is plainly different from that of the robbers who killed; yet the State treated them alike and attributed to Enmund the culpability of those who killed the Kerseys. This was impermissible under the Eighth Amendment.” Ibid, (emphasis added).
Enmund obviously cast considerable doubt on the constitutionality of the death sentences imposed on petitioners in this case. Following the Enmund decision, petitioners applied to the Arizona Supreme Court for postconviction review. They argued that Enmund prevented the State from imposing the death sentence because they, like Enmund, were accomplices to a felony in which killings occurred that they neither committed nor intended to commit. Despite its earlier holding that petitioners had not killed or intended to kill anyone, the Arizona Supreme Court again upheld the Tisons’ sentences. First, the court defined intent broadly, adopting a definition that equates “intent to kill” with the foreseeability of harm:
“Intend [sic] to kill includes the situation in which the defendant intended, contemplated, or anticipated that lethal force would or might be used or that life would or might be taken in accomplishing the underlying felony.” 142 Ariz. 454, 456, 690 P. 2d 755, 757 (1984).
The court then reviewed, in a passage this Court quotes at length, ante, at 144-145, petitioners’ conduct during the *163escape and subsequent flight. The court did not attempt to link any of petitioners’ statements or actions to the decision to kill the family, nor did it make any findings concerning petitioners’ mental states at the time of the shootings. Instead, the court found that each petitioner “could [have] anticipate^] the use of lethal force during this attempt to flee confinement.” 142 Ariz. 446, 448, 690 P. 2d 747, 749 (1984); 142 Ariz., at 456, 690 P. 2d, at 757. The Arizona Supreme Court then held, by a vote of 3-2, that this finding was sufficient to establish that petitioners “intended” (within the meaning of Enmund) to kill the Lyons family, and affirmed the death sentences.
The Arizona Supreme Court thus attempted to comply with Enmund by making a finding as to petitioners’ mental state. The foreseeability standard that the court applied was erroneous, however, because “the possibility of bloodshed is inherent in the commission of any violent felony and this possibility is generally foreseeable and foreseen.” Ante, at 151. Under the lower court’s standard, any participant in a violent felony during which a killing occurred, including Enmund, would be liable for the death penalty. This Court therefore properly rejects today the lower court’s misguided attempt to preserve its earlier judgment by equating intent with foreseeable harm. Ante, at 150-151. In my view, this rejection completes the analytic work necessary to decide this case, and on this basis petitioners’ sentences should have been vacated and the judgment reversed.
The Court has chosen instead to announce a new substantive standard for capital liability: a defendant’s “major participation in the felony committed, combined with reckless indifference to human life, is sufficient to satisfy the Enmund culpability requirement.” Ante, at 158. The Court then remands the case for a determination by the state court whether petitioners are culpable under this new standard. Nevertheless, the Court observes, in dictum, that “the record would support a finding of the culpable mental state of *164reckless indifference to human life.” Ante, at 151; see also ante, at 152. (“These facts . . . would clearly support a finding that [both sons] subjectively appreciated that their acts were likely to result in the taking of innocent life”).
I join no part of this. First, the Court’s dictum that its new category of mens rea is applicable to these petitioners is not supported by the record. Second, even assuming petitioners may be so categorized, objective evidence and this Court’s Eighth Amendment jurisprudence demonstrate that the death penalty is disproportionate punishment for this category of defendants. Finally, the fact that the Court reaches a different conclusion is illustrative of the profound problems that continue to plague capital sentencing.
HH HH
The facts on which the Court relies are not sufficient, in my view, to support the Court’s conclusion that petitioners acted with reckless disregard for human life.4 But even if they *165were, the Court’s decision to restrict its vision to the limited set of facts that “the Arizona Supreme Court has given . . . to us,” ante, at 151, is improper.5 By limiting itself to the facts the lower court found relevant to the foreseeability standard, this Court insulates itself from other evidence in the record directly relevant to the new standard articulated today. This evidence suggests that the question of petitioners’ mental states with respect to the shootings is very much an open one to be decided only after a thorough evidentiary hearing. I therefore stress that nothing in the Court’s opinion abrogates the State’s responsibility independently and fairly to consider all the relevant evidence before applying the Court’s new standard. See Cabana v. Bullock, 474 U. S. 376, 391 (1986) (“Considerations of federalism and comity counsel respect for the ability of state courts to carry out their role as the primary protectors of the rights of criminal defendants”).
The evidence in the record overlooked today regarding petitioners’ mental states with respect to the shootings is not trivial. For example, while the Court has found that petitioners made no effort prior to the shooting to assist the victims, the uncontradicted statements of both petitioners are *166that just prior to the shootings they were attempting to find a jug of water to give to the family. App. 20-21, 39-41, 74-75, 109. While the Court states that petitioners were on the scene during the shooting and that they watched it occur, Raymond stated that he and Ricky were still engaged in repacking the Mazda after finding the water jug when the shootings occurred. Id., at 21, 75. Ricky stated that they had returned with the water, but were still some distance (“farther than this room”) from the Lincoln when the shootings started, id., at 40-41, 111, and that the brothers then turned away from the scene and went back to the Mazda, id., at 113. Neither stated that they anticipated that the shootings would occur, or that they could have done anything to prevent them or to help the victims afterward.6 Both, however, expressed feelings of surprise, helplessness, and regret. This statement of Raymond’s is illustrative:
“Well, I just think you should know when we first came into this we had an agreement with my dad that nobody would get hurt because we [the brothers] wanted no one hurt. And when this [killing of the kidnap victims] came about we were not expecting it. And it took us by surprise as much as it took the family [the victims] by surprise because we were not expecting this to happen. And I feel bad about it happening. I wish we could [have done] something to stop it, but by the time it happened it was too late to stop it. And it’s just something *167we are going to live with the rest of our lives. It will always be there.” 142 Ariz., at 462, 690 P. 2d, at 763; see also App. 242.7
*168In light of this evidence, it is not surprising that the Arizona Supreme Court rested its judgment on the narrow ground that petitioners could have anticipated that lethal force might be used during the escape, or that the state probation officer — who reviewed at length all the facts concerning the sons’ mental states — did not recommend that the death sentence be imposed. The discrepancy between those aspects of the record on which the Court has chosen to focus and those aspects it has chosen to ignore underscores the point that a reliable and individualized Enmund determination can be made only by the trial court following an eviden-tiary hearing. See Cabana v. Bullock, 474 U. S., at 397-407 (Blackmun, J., dissenting); id., at 407-408 (Stevens, J., dissenting).
Ill
Notwithstanding the Court’s unwarranted observations on the applicability of its new standard to this case, the basic flaw in today’s decision is the Court’s failure to conduct the sort of proportionality analysis that the Constitution and past cases require. Creation of a new category of culpability is not enough to distinguish this case from Enmund. The Court must also establish that death is a proportionate punishment for individuals in this category. In other words, the Court must demonstrate that major participation in a felony with a state of mind of reckless indifference to human life deserves the same punishment as intending to commit a murder or actually committing a murder. The Court does not attempt to conduct a proportionality review of the kind performed in past cases raising a proportionality question, e. g., Solem v. Helm, 463 U. S. 277 (1983); Enmund v. Florida, 458 U. S. 782 (1982); Coker v. Georgia, 433 U. S. 584 (1977), but instead offers two reasons in support of its view.
A
One reason the Court offers for its conclusion that death is proportionate punishment for persons falling within its new *169category is that limiting the death penalty to those who intend to kill “is a highly unsatisfactory means of definitively distinguishing the most culpable and dangerous of murderers.” Ante, at 157. To illustrate that intention cannot be dispositive, the Court offers as examples “the person who tortures another not caring whether the victim lives or dies, or the robber who shoots someone in the course of the robbery, utterly indifferent to the fact that the desire to rob may have the unintended consequence of killing the victim as well as taking the victim’s property.” Ibid, (emphasis added). Influential commentators and some States have approved the use of the death penalty for persons, like those given in the Court’s examples, who kill others in circumstances manifesting an extreme indifference to the value of human life.8 Thus an exception to the requirement that only intentional murders be punished with death might be made for persons who actually commit an act of homicide; Enmund, by distinguishing from the accomplice case “those who kill,” clearly reserved that question. But the constitutionality of the death penalty for those individuals is no more relevant to this case than it was to Enmund, because this case, like Enmund, involves accomplices who did not kill. Thus, although some of the “most culpable and dangerous of murderers” may be those who killed without specifically intending to kill, it is considerably more difficult to apply that rubric con*170vincingly to those who not only did not intend to kill, but who also have not killed.9
It is precisely in this context — where the defendant has not killed — that a finding that he or she nevertheless intended to kill seems indispensable to establishing capital culpability. It is important first to note that such a defendant has not committed an act for which he or she could be sentenced to death. The applicability of the death penalty therefore turns entirely on the defendant’s mental state with regard to an act committed by another. Factors such as the defendant’s major participation in the events surrounding the killing or the defendant’s presence at the scene are relevant insofar as they illuminate the defendant’s mental state with regard to the killings. They cannot serve, however, as independent grounds for imposing the death penalty.
Second, when evaluating such a defendant’s mental state, a determination that the defendant acted with intent is qualitatively different from a determination that the defendant acted with reckless indifference to human life. The difference lies in the nature of the choice each has made. The reckless actor has not chosen to bring about the killing in the way the intentional actor has. The person who chooses to *171act recklessly and is indifferent to the possibility of fatal consequences often deserves serious punishment. But because that person has not chosen to kill, his or her moral and criminal culpability is of a different degree than that of one who killed or intended to kill.
The importance of distinguishing between these different choices is rooted in our belief in the “freedom of the human will and a consequent ability and duty of the normal individual to choose between good and evil.” Morissette v. United States, 342 U. S. 246, 250 (1952). To be faithful to this belief, which is “universal and persistent in mature systems of law,” ibid., the criminal law must ensure that the punishment an individual receives conforms to the choices that individual has made.10 Differential punishment of reckless and intentional actions is therefore essential if we are to retain “the relation between criminal liability and moral culpability” on which criminal justice depends. People v. Washington, 62 Cal. 2d 777, 783, 402 P. 2d 130, 134 (1965) (opinion of Traynor, C. J.). The State’s ultimate sanction — if it is ever to be used — must be reserved for those whose culpability is greatest. Cf. Enmund, 458 U. S., at 798 (“It is fundamental that ‘causing harm intentionally must be punished more severely than causing the same harm unintentionally’ ” (citation omitted)); United States v. United States Gypsum Co., 438 U. S. 422, 444 (1978).
Distinguishing intentional from reckless action in assessing culpability is particularly important in felony-murder cases. Justice White stressed the importance of this distinction in Lockett v. Ohio, 438 U. S. 586 (1978), a felony-murder case in *172which the petitioner’s death sentence was vacated on other grounds.
“[S]ociety has made a judgment, which has deep roots in the history of the criminal law. . . distinguishing at least for purpose of the imposition of the death penalty between the culpability of those who acted with and those who acted without a purpose to destroy life.
“[T]he type of conduct which Ohio would punish by death requires at most the degree of mens rea defined by the ALI Model Penal Code (1962) as recklessness: conduct undertaken with knowledge that death is likely to follow. Since I would hold that death may not be inflicted for killings consistent with the Eighth Amendment without a finding that the defendant engaged in conduct with the conscious purpose of producing death, these sentences must be set aside.” Id., at 626-628 (emphasis added; footnotes omitted).
In Enmund, the Court explained at length the reasons a finding of intent is a necessary prerequisite to the imposition of the death penalty. In any given case, the Court said, the death penalty must “measurably contribute]” to one or both of the two “social purposes” — deterrence and retribution— which this Court has accepted as justifications for the death penalty. Enmund, supra, at 798, citing Gregg v. Georgia, 428 U. S. 153, 183 (1976). If it does not so contribute, it “‘is nothing more than the purposeless and needless imposition of pain and suffering’ and hence an unconstitutional punishment.” Enmund, supra, at 798, quoting Coker v. Georgia, 433 U. S., at 592. Enmund’s lack of intent to commit the murder — rather than the lack of evidence as to his mental state — was the decisive factor in the Court’s decision that the death penalty served neither of the two purposes. With regard to deterrence, the Court was
*173“quite unconvinced . . . that the threat that the death penalty will be imposed for murder will measurably deter one who does not kill and has no intention or purpose that life wall be taken. Instead, it seems likely that ‘capital punishment can serve as a deterrent only when murder is the result of premeditation and deliberation’ . . . Enmund, supra, at 798-799.11
As for retribution, the Court again found that Enmund’s lack of intent, together with the fact that he did not kill the victims, was decisive. “American criminal law has long considered a defendant’s intention — and therefore his moral guilt— to be critical to the ‘degree of [his] criminal culpability.’” 458 U. S., at 800 (citation omitted). The Court concluded that “[pjutting Enmund to death to avenge two killings that he did not commit and had no intention of committing or causing does not measurably contribute to the retributive end of ensuring that the criminal gets his just deserts. ” Id., at 801. Thus, in Enmund the Court established that a finding of an intent to kill was a constitutional prerequisite for the imposition of the death penalty on an accomplice who did not kill. The Court has since reiterated that “Enmund . . . imposes a categorical rule: a person who has not in fact killed, attempted to kill, or intended that a killing take place or that lethal force be used may not be sentenced to death.” Ca*174bana v. Bullock, 474 U. S., at 386. The Court’s decision today to approve the death penalty for accomplices who lack this mental state is inconsistent with Enmund and with the only justifications this Court has put forth for imposing the death penalty in any case.
B
The Court’s second reason for abandoning the intent requirement is based on its survey of state statutes authorizing the death penalty for felony murder, and on a handful of state cases.12 On this basis, the Court concludes that “[o]nly *175a small minority of those jurisdictions imposing capital punishment for felony murder have rejected the possibility of a capital sentence absent an intent to kill, and we do not find this minority position constitutionally required.” Ante, at 158 (emphasis added). The Court would thus have us believe that “the majority of American jurisdictions clearly authorize capital punishment” in cases such as this. Ante, at 155. This is not the case. First, the Court excludes from its survey those jurisdictions that have abolished the death penalty and those that have authorized it only in circumstances different from those presented here. When these jurisdictions are included, and are considered with those jurisdictions that require a finding of intent to kill in order to impose the death sentence for felony murder, one discovers that approximately three-fifths of American jurisdictions do not authorize the death penalty for a nontriggerman absent a finding that he intended to kill. Thus, contrary to the Court’s implication that its view is consonant with that of “the majority of American jurisdictions,” ibid., the Court’s view is itself distinctly the minority position.13
*176Second, it is critical to examine not simply those jurisdictions that authorize the death penalty in a given circumstance, but those that actually impose it. Evidence that a penalty is imposed only infrequently suggests not only that jurisdictions are reluctant to apply it but also that, when it is applied, its imposition is arbitrary and therefore unconstitutional. Furman v. Georgia, 408 U. S. 238 (1972). Thus, the Court in Enmund examined the relevant statistics on the imposition of the death penalty for accomplices in a felony murder. The Court found that of all executions between 1954 and 1982, there were “only 6 cases out of 362 where a nontriggerman felony murderer was executed. All six executions took place in 1955 ” 458 U. S., at 794 (emphasis added). This evidence obviously militates against imposing the death penalty on petitioners as powerfully as it did against imposing it on Enmund.14
*177The Court in Enmund also looked at the imposition of the death penalty for felony murder within Florida, the State that had sentenced Enmund. Of the 45 murderers then on death row, 36 had been found to have “intended” to take life, and 8 of the 9 for which there was no finding of intent had been the triggerman. Thus in only one case — Enmund— had someone (such as the Tisons) who had neither killed nor intended to kill received the death sentence. Finally, the Court noted that in no Commonwealth or European country could Enmund have been executed, since all have either abolished or never employed a felony-murder doctrine. Id., at 796-797, n. 22.15
The Court today neither reviews nor updates this evidence. Had it done so, it would have discovered that, even *178including the 65 executions since Enmund, “[t]he fact remains that we are not aware of a single person convicted of felony murder over the past quarter century who did not kill or attempt to kill, and did not intend the death of the victim, who has been executed . . . 458 U. S., at 796.16 Of the 64 persons on death row in Arizona, all of those who have raised and lost an Enmund challenge in the Arizona Supreme Court have been found either to have killed or to have specifically intended to kill.17 Thus, like Enmund, the Tisons’ sentence *179appears to be an aberration within Arizona itself as well as nationally and internationally. The Court’s objective evidence that the statutes of roughly 20 States appear to authorize the death penalty for defendants in the Court’s new category is therefore an inadequate substitute for a proper proportionality analysis, and is not persuasive evidence that the punishment that was unconstitutional for Enmund is constitutional for the Tisons.
C
The Court’s failure to examine the full range of relevant evidence is troubling not simply because of what that examination would have revealed, but because until today such an examination has been treated as constitutionally required whenever the Court undertakes to determine whether a given punishment is disproportionate to the severity of a given crime. Enmund is only one of a series of cases that have framed the proportionality inquiry in this way. See, e. g., Coker v. Georgia, 433 U. S. 584 (1977). In the most recent stich case, Solem v. Helm, 463 U. S. 277, 292 (1983), the Court summarized the essence of the inquiry:
“In sum, a court’s proportionality analysis under the Eighth Amendment should be guided by objective criteria, including (i) the gravity of the offense and the harshness of the penalty; (ii) the sentences imposed on other criminals in the same jurisdiction; and (iii) the sentences *180imposed for commission of the same crime in other jurisdictions.” (Emphasis added.)
By addressing at best only the first of these criteria, the Court has ignored most of the guidance this Court has developed for evaluating the proportionality of punishment.
Such guidance is essential in determining the constitutional limits on the State’s power to punish. These limits must be defined with care, not simply because the death penalty is involved, but because the social purposes that the Court has said justify the death penalty — retribution and deterrence-are justifications that possess inadequate self-limiting principles. As Professor Packer observed, under a theory of deterrence the state may justify such punishments as “boiling people in oil; a slow and painful death may be thought more of a deterrent to crime than a quick and painless one.” Packer, Making the Punishment Fit the Crime, 77 Harv. L. Rev. 1071, 1076 (1964).18 Retribution, which has as its core logic *181the crude proportionality of “an eye for an eye,” has been regarded as a constitutionally valid basis for punishment only when the punishment is consistent with an “individualized consideration” of the defendant’s culpability, Lockett v. Ohio, 438 U. S., at 605, and when “the administration of criminal justice” works to “channel!]” society’s “instinct for retribution.” Furman v. Georgia, 408 U. S., at 308 (Stewart, J., concurring). Without such channeling, a State could impose a judgment of execution by torture as appropriate retribution for murder by torture.19 Thus, under a simple theory either of deterrence or retribution, unfettered by the Constitution, results disturbing to civil sensibilities and inconsistent with “the evolving standards of decency” in our society become rationally defensible. Cf. Trop v. Dulles, 356 U. S. 86, 101 (1958).
The Framers provided in the Eighth Amendment the limiting principles otherwise absent in the prevailing theories of punishment. One such principle is that the States may not impose punishment that is disproportionate to the severity of *182the offense or to the individual’s own conduct and culpability. Because the proportionality inquiry in this case overlooked evidence and considerations essential to such an inquiry, it is not surprising that the result appears incongruous. Ricky and Raymond Tison are similarly situated with Earl Enmund in every respect that mattered to the decision in Enmund. Like Enmund, the Tisons neither killed nor attempted or intended to kill anyone. Like Enmund, the Tisons have been sentenced to death for the intentional acts of others which the Tisons did not expect, which were not essential to the felony, and over which they had no control. Unlike Enmund, however, the Tisons will be the first individuals in over 30 years to be executed for such behavior.
I conclude that the proportionality analysis and result in this case cannot be reconciled with the analyses and results of previous cases. On this ground alone, I would dissent. But the fact that this Court’s death penalty jurisprudence can validate different results in analytically indistinguishable cases suggests that something more profoundly disturbing than faithlessness to precedent is at work in capital sentencing.
f — I <1
In 1922, “five negroes who were convicted of murder m the first degree and sentenced to death by the Court of the State of Arkansas” appealed to this Court from an order of the District Court dismissing their writ of habeas corpus. Moore v. Dempsey, 261 U. S. 86, 87 (1923). The crux of their appeal was that they “were hurried to conviction under the pressure of a mob without any regard for their rights and without according to them due process of law.” Ibid. In reversing the order, Justice Holmes stated the following for the Court:
“It certainly is true that mere mistakes of law in the course of a trial are not to be corrected [by habeas corpus]. But if the case is that the whole proceeding is a mask — that counsel, jury, and judge were swept to the fatal end by an irresistible wave of public passion, and *183that the State Courts failed to correct the wrong, neither perfection in the machinery for correction nor the possibility that the trial court and counsel saw no other way of avoiding an immediate outbreak of the mob can prevent this Court from securing to the petitioners their constitutional rights.” Id., at 91.
A
In Furman v. Georgia, supra, this Court concluded that the State’s procedural machinery was so imperfect that imposition of the death penalty had become arbitrary and therefore unconstitutional. A scant four years later, however, the Court validated Georgia’s new machinery, and in 1977 executions resumed. In this case, the State appears to have afforded petitioners all of the procedures that this Court has deemed sufficient to produce constitutional sentencing decisions. Yet in this case, as in Moore, “perfection in the [State’s] machinery for correction” has not secured to petitioners their constitutional rights. So rarely does any State (let alone any Western country other than our own) ever execute a person who neither killed nor intended to kill that “these death sentences are cruel and unusual in the same way that being struck by lightning is cruel and unusual.” Furman v. Georgia, supra, at 309 (Stewart, J., concurring). This case thus demonstrates, as Furman also did, that we have yet to achieve a system capable of “distinguishing the few cases in which the [death penalty] is imposed from the many cases in which it is not.” 408 U. S., at 313 (White, J., concurring).
What makes this a difficult case is the challenge of giving substantive content to the concept of criminal culpability. Our Constitution demands that the sentencing decision itself, and not merely the procedures that produce it, respond to the reasonable goals of punishment. But the decision to execute these petitioners, like the state courts’ decisions in Moore, and like other decisions to kill, appears responsive *184less to reason than to other, more visceral, demands. The urge to employ the felony-murder doctrine against accomplices is undoubtedly strong when the killings stir public passion and the actual murderer is beyond human grasp. And an intuition that sons and daughters must sometimes be punished for the sins of the father may be deeply rooted in our consciousness.20 Yet punishment that conforms more closely to such retributive instincts than to the Eighth Amendment is tragically anachronistic in a society governed by our Constitution.
B
This case thus illustrates the enduring truth of Justice Harlan’s observation that the tasks of identifying “those characteristics of criminal homicides and their perpetrators which call for the death penalty, and [of] expressing] these characteristics in language which can be fairly understood and applied by the sentencing authority appear to be . . . beyond present human ability.” McGautha v. California, 402 U. S. 183, 204 (1971) (emphasis added). The persistence of doctrines (such as felony murder) that allow excessive discretion in apportioning criminal culpability and of decisions (such as today’s) that do not even attempt “precisely [to] delineate the particular types of conduct and states of mind warranting imposition of the death penalty,” ante, at 158, demonstrates that this Court has still not articulated rules that will ensure that capital sentencing decisions conform to the substantive principles of the Eighth Amendment. Arbitrariness continues so to infect both the procedure and substance of capital sentencing that any decision to impose the *185death penalty remains cruel and unusual. For this reason, as well as for the reasons expressed in Gregg v. Georgia, 428 U. S., at 227, 1 adhere to my view that the death penalty is in all circumstances cruel and unusual punishment prohibited by the Eighth and Fourteenth Amendments, and dissent.
App. 297 (quoting Paul Dean in the Arizona Republic, Aug. 16, 1978).
As explained in the Commentaries on the Model Penal Code: “At common law all felonies were punishable by death. In a felony-murder situation, it made little difference whether the actor was convicted of murder or of the underlying felony because the sanction was the same. The primary use of the felony-murder rule at common law therefore was to deal with a homicide that occurred in furtherance of an attempted felony that failed. Since attempts were punished as misdemeanors, . . . the use of the felony-*160murder rule allowed the courts to punish the actor in the same manner as if his attempt had succeeded. Thus, a conviction for attempted robbery was a misdemeanor, but a homicide committed in the attempt was murder and punishable by death.” ALI, Model Penal Code Commentaries §210.2, p. 31, n. 74 (Off. Draft 1980).
As the Court notes, ante, at 146, n. 2, it has expressed no view on the constitutionality of Arizona’s decision to attribute to petitioners as an aggravating factor the manner in which other individuals carried out the killings. On its face, however, that decision would seem to violate the core Eighth Amendment requirement that capital punishment be based on an “individualized consideration” of the defendant’s culpability, Lockett v. *161Ohio, 438 U. S. 586, 605 (1978). It therefore remains open to the state courts to consider whether Arizona’s aggravating factors were interpreted and applied so broadly as to violate the Constitution. Godfrey v. Georgia, 446 U. S. 420 (1980).
Petitioners’ presence at the scene of the murders, and their participation in flagging down the vehicle, and robbing and guarding the family, indicate nothing whatsoever about their subjective appreciation that their father and his friend would suddenly decide to kill the family. Each of petitioners’ actions was perfectly consistent with, and indeed necessary to, the felony of stealing a car in order to continue the flight from prison. Nothing in the record suggests that any of their actions were inconsistent with that aim. Indeed, the trial court recognized the disjunction between the felonies and the murders when it found that Gary Tison’s and Greenawalt’s decision to murder the family was senseless and unnecessary to the escape. The court based its finding of aggravating circumstances in part “on the senselessness of the murders,” and stated that:
“It was not essential to the defendants’ continuing evasion of arrest that these persons were murdered. The victims could easily have been restrained sufficiently to permit the defendants to travel a long distance before the robberies, the kidnappings, and the theft were reported.” App. 283.
Thus the Court’s findings about petitioners’ mental states regarding the murders are based solely on inferences from petitioners’ participation in the underlying felonies. Their decision to provide arms for and participate *165in a prison breakout and escape may support the lower court’s finding that they should have anticipated that lethal force might be used during the breakout and subsequent flight, but it does not support the Court’s conclusions about petitioners’ mental states concerning the shootings that actually occurred.
When the Arizona Supreme Court first reviewed this case on appeal, it stated that petitioners’ degree of mens rea was of little significance to the case. On rehearing, the Arizona Supreme Court did make a finding that petitioners could have anticipated that lethal force would be used during the breakout or subsequent flight. In that regard, it referred to facts concerning thé breakout and escape. See ante, at 143-145. The court did not refer to the evidence in the record of petitioners’ mental states concerning the actual shootings, however, nor was such evidence relevant to its decision. Given the question it had chosen to address, evidence regarding petitioners’ actual mental states with regard to the shooting was superfluous.
In addition, the Court’s statement that Raymond did not act to assist the victims “after” the shooting, and its statement that Ricky “watched the killing after which he chose to aid those whom he had placed in the position to kill rather than their victims,” ante, at 162, takes license with the facts found by the Arizona Supreme Court. That court did not say whether petitioners did anything to help the victims following the shooting, nor did it make any findings that would lead one to believe that something could have been done to assist them. The lower court merely stated that petitioners did not “disassociate” themselves from their father and Greenawalt after the shooting. Ante, at 145 (citation omitted).
These expressions are consistent with other evidence about the sons’ mental states that this Court, like the lower courts, has neglected. Neither son had a prior felony record. App. 233-234. Both lived at home with their mother, and visited their father, whom they believed to be “a model prisoner,” each week. See Brief for Petitioners 3 (citing Tr. of Mar. 14,1979, hearing). They did not plan the breakout or escape; rather their father, after thinking about it himself for a year, mentioned the idea to Raymond for the first time one week before the breakout, and discussed with his sons the possibility of having them participate only the day before the breakout. App. 50-51, 91. The sons conditioned their participation on their father’s promise that no one would get hurt; during the breakout, their father kept his word. The trial court found that the murders then-father later committed were senseless and unnecessary to the felony of stealing a car in which the sons participated; and just prior to the shootings the sons were retrieving a water jug for the family. Given these circumstances, the sons’ own testimony that they were surprised by the killings, and did not expect them to occur, appears more plausible than the Court’s speculation that they “subjectively appreciated that their activities were likely to result in the taking of innocent life.” Ante, at 152. The report of the psychologist, who examined both sons, also suggests that they may not have appreciated the consequences of their participation:
“These most unfortunate youngsters were bom into an extremely pathological family and were exposed to one of the premier sociopaths of recent Arizona history. In my opinion this very fact had a severe influence upon the personality structure of these youngsters ....
“I do believe that their father, Gary Tison, exerted a strong, consistent, destructive but subtle pressure upon these youngsters and I believe that these young men got committed to an act which was essentially ‘over their heads.’ Once committed, it was too late and there does not appear to be any true defense based on brainwashing, mental deficiency, mental illness or irresistable urge. There was a family obsession, the boys were ‘trained’ to think of their father as an innocent person being victimized in the state prison but both youngsters have made perfectly clear that they were functioning of their own volition. At a deeper psychological level it may have been less of their own volition than as a result of Mr. Tison’s ‘conditioning’ and the rather amoral attitudes within the family home.” Brief for Petitioners 11-12, n. 16.
For example, the Court quotes Professor Fletcher’s observation that “the Model Penal Code treats reckless killing ... as equivalent to purposeful and knowing killing.” Ante, at 157 (emphasis added). The Model Penal Code advocates replacing the felony-murder rule with a rule that allows a conviction for murder only when the killer acted with intent, purpose, or “reckless[ness] under circumstances manifesting extreme indifference to the value of human life.” See ALI, Model Penal Code Commentaries §210.2, p. 13 (Off. Draft 1980). The Code offers as examples shooting into a crowd or an automobile, or shooting a person in the course of playing Russian roulette. Id., at 22-23.
A second problem with the Court’s examples is that they illustrate wanton, but nevertheless intentional, killings, rather than unintentional killings. The element that these wanton killings lack is not intent, but rather premeditation and deliberation. Professor Fletcher explains the point:
“[Wjhile planning and calculation represent one form of heinous or coldblooded murder, premeditation is not the only feature that makes intentional killings wicked. Wanton killings are generally regarded as among the most wicked, and the feature that makes a killing wanton is precisely the absence of detached reflection before the deed. Fitzjames Stephen put the case of a man who ‘sees a boy sitting on a bridge over a deep river and, out of mere wanton barbarity, pushes him into it and so drowns him.’ Killing without a motive can usually be just as wicked as killing after detached reflection about one’s goals.” G. Fletcher, Rethinking Criminal Law 254 (1978) (footnote omitted; emphasis added).
We show this fidelity, for example, when we decline to hold a young child as morally and criminally responsible for an illegal act as we would hold an adult who committed the same act. Although the child has committed the illegal act and caused the harmful result, the child’s actions are presumed not to reflect a mature capacity for choice, and the child’s culpability for the act is accordingly reduced.
The Court acknowledged that “[i]t would be very different if the likelihood of a killing in the course of a robbery were so substantial that one should share the blame for the killing if he somehow participated in the felony.” 458 U. S., at 799. Nevertheless, the Court saw no reason to depart from its conclusion that the death penalty could not be justified as a deterrent in that case, because “competent observers have concluded that there is no basis in experience for the notion that death so frequently occurs in the course of a felony for which killing is not an essential ingredient that the death penalty should be considered as a justifiable deterrent to the felony itself.” Ibid. The trial court found that the killings in the case were not an essential ingredient of the felony. App. 283, quoted infra, at 164, n. 4. Thus the goal of deterrence is no more served in this case than it was in Enmund.
We should be reluctant to conclude too much from the Court’s survey of state decisions, because most jurisdictions would not approve the death penalty in the circumstances here, see n. 13, infra, and the Court neglects decisions applying the law of those States. E. g., Clark v. Louisiana State Penitentiary, 694 F. 2d 75 (CA5 1982) (under Louisiana law, jury must find specific intent to kill); People v. Garcia, 36 Cal. 3d 539, 684 P. 2d 826 (1984) (death penalty for felony murder may not be imposed without finding of specific intent to kill), cert. denied, 469 U. S. 1229 (1985).
Moreover, the cases the Court does cite are distinguishable from this ease. In four of the five cases cited as evidence of an “apparent consensus” that intent to kill is not a prerequisite for imposing the death penalty, the court did not specifically find an absence of any act or intent to kill. Moreover, in each of these cases the court at least suggested that the defendants intended to kill, attempted to kill, or participated in the actual killing. Clines v. State, 280 Ark. 77, 84, 656 S. W. 2d 684, 687 (1983) (“There was direct evidence from more than one source that appellants had discussed among themselves the necessity of murder if they met resistance” and evidence that victim “was immediately attacked by appellants, sustaining blows to his head and face from the metal chain and a mortal wound to the chest”), cert. denied, 465 U. S. 1051 (1984); Deputy v. State, 500 A. 2d 581, 599 (Del. 1985) (“Deputy was not solely a participant in the underlying felony, but was instead present during, and involved in, the actual murders”), cert. pending, No. 85-6272; Ruffin v. State, 420 So. 2d 591, 594 (Fla. 1982) (“Evidence is abundantly clear and sufficient to demonstrate Ruffin’s joint participation in the premeditated murder of Karol Hurst”); Selvage v. State, 680 S. W. 2d 17, 22 (Tex. Cr. App. 1984) (“Unlike Enmund, appellant used lethal force to effectuate a safe escape and attempted to kill Ventura and Roberts as they pursued him and his companion from the jewelry store”). As for the fifth case, People v. Davis, 95 Ill. 2d 1, 52-53, 447 N. E. 2d 353, 378-379 (1983) (defendant received death *175sentence for his role in successive burglaries during each of which codefen-dant killed resident), the court appears to have held that the defendant “knew” that his codefendant would commit the murder, a mental state significantly different than that attributed to the Tisons.
Thirteen States and the District of Columbia have abolished the death penalty. NAACP Legal Defense and Educational Fund, Death Row U. S. A. 1 (Aug. 1986). According to the Court, ante, at 154-156, n. 10, 11 States would not authorize the death penalty in the circumstances presented here. At least four other States not cataloged by the Court also restrict the imposition of capital punishment to those who actually commit and intend to commit murder, and two more States reject the death penalty for most felony murders, see this note infra, at 176. In addition, the Supreme Court of at least one of the States cited by the majority as a State authorizing the death penalty absent a finding of intent has explicitly ruled that juries must find that a felony-murder defendant had a specific intent to kill before imposing the death sentence. Carlos v. Superior Court of Los Angeles Co., 35 Cal. 3d 131, 672 P. 2d 862 (1983). Thus it appears that *176about three-fifths of the States and the District of Columbia have rejected the position the Court adopts today.
For States that restrict the imposition of capital punishment to those who actually and intentionally kill, see Mo. Rev. Stat. §§ 565.001, 565.003, 565.020 (1986) (death penalty reserved for those who intentionally, knowingly, and deliberately cause death); 18 Pa. Cons. Stat. §§ 2502(a), (b), (d), 1102 (1982) (death penalty reserved for those who commit an intentional killing); Vt. Stat. Ann., Tit. 13, §§2303(b), (c) (Supp. 1986) (only murderers of correctional officers subject to death penalty); Wash. Rev. Code §§9A.32.030,10.95.020 (1985) (death penalty reserved for those who commit premeditated killing with at least one aggravating circumstance). Two other States also forbid imposition of the death penalty under the general standards announced today, although other aspects of their statutes might render them applicable to these defendants on the facts of this case. See Md. Ann. Code, Art. 27, §§410, 412(b), 413(d)(10), 413(e)(1), 413(d)(5) (1957 and Supp. 1986) (death penalty may be imposed only on person who committed the killing, but possible exception if victim is a child); N. H. Rev. Stat. Ann. §§ 630:1, 630:1(111), 630:1-a(I)(b)(2) (1986) (death penalty reserved for killing a law enforcement officer, murder for hire, and killing during a kidnaping).
Although the Court ignores the statistics on actual executions, it does refer earlier in its opinion to the evidence discussed in Enmund that of the 739 inmates on death row for whom sufficient data were available, only 41 *177did not participate in the fatal assault on the victim and only 16 were not present. Ante, at 148; see Enmund, 468 U. S., at 796. While in En-mund the Court focused on a breakdown of these statistics into those physically present at the scene and those not, that information is not relevant here. What would be relevant, and what the summary in Enmund does not tell us, is how many of the 41 who did not participate were also found not to have intended that the murder occur.
Although statistics on the average sentences given for nontriggermen in felony murders were not presented to the Court, it is possible that such statistics would reveal a wide range of results. One felony-murder case worth noting in this regard is People v. Ganter, 56 Ill. App. 3d 316, 371 N. E. 2d 1072 (1977). Ganter and a codefendant committed an armed robbery of a store, during which Ganter killed one of the store’s owners. “The evidence at trial showed defendant was the actual murderer. He shot Thomas at close range, without provocation and as Thomas stood in a helpless position. The accomplice, although accountable for the death by his participation in the attempt [sic] armed robbery, did not do the actual killing.” Id., at 328, 371 N. E. 2d, at 1080-1081. Ganter was sentenced to 20-30 years; his accomplice was sentenced to 3-6 years. Id., at 321, 327, 371 N. E. 2d, at 1076, 1080.
Since Enmund was decided, the Netherlands and Australia have abolished the death penalty for all offenses, and Cyprus, El Salvador, and Argentina have abolished it for all crimes except those committed in wartime or in violation of military law. Amnesty International, United States of America, The Death Penalty 228-231 (1987).
Lists of those executed and those on death row are published in NAACP Legal Defense Fund, Death Row U. S. A. (Mar. 1987). Review of those executed since 1982 reveals that each person executed was found to have committed a killing and/or to have intended to kill. In only two eases does there remain some doubt whether the person executed actually killed the victim; in each case, however, the defendant was found at a minimum to have intended to kill. Green v. Zant, 738 F. 2d 1529, 1533-1534 (CA11) (case was presented to jury on malice-murder rather than felony-murder theory, and evidence supported verdict on that theory), cert. denied, 469 U. S. 1098 (1984); Skillern v. Estelle, 720 F. 2d 839, 844 (CA5 1983) (evidence supports finding that Skillern agreed and “plotted in advance” to kill the eventual victim), cert, denied sub nom. Skillern v. Procunier, 469 U. S. 1067 (1984).
See Amnesty International, swpra, at 192 (listing death row totals by State as of Oct. 1986). The cases since Enmund in which the Arizona Supreme Court has rejected the defendant’s Enmund challenge and affirmed the death sentence are: State v. Correll, 148 Ariz. 468, 478, 715 P. 2d 721, 731 (1986) (defendant intended to kill victims and “verbally encouraged” codefendant to proceed with killing); State v. Martinez-Villareal, 145 Ariz. 441, 702 P. 2d 670 (defendant actively took part in the murder and intended to kill), cert. denied, 474 U. S. 975 (1985); State v. Hooper, 145 Ariz. 538, 703 P. 2d 482 (1985) (defendant killed for hire), cert. denied, 474 U. S. 1073 (1986); State v. Bishop, 144 Ariz. 521, 698 P. 2d 1240 (1985) (defendant planned and intended to kill, assaulted victim, and abandoned victim in mine shaft); State v. Poland, 144 Ariz. 388, 698 P. 2d 183 (1985) (defendants killed victims), aff’d, 476 U. S. 147 (1986); State v. Villafuerte, 142 Ariz. 323, 690 P. 2d 42 (1984) (defendant killed victim), cert. denied, 469 U. S. 1230 (1985); State v. Fisher, 141 Ariz. 227, 686 P. 2d 750 (defendant killed victim), cert. denied, 469 U. S. 1066 (1984); State v. James, 141 Ariz. 141, 685 P. 2d 1293 (defendant killed and intended to kill), cert. denied, 469 U. S. 990 (1984); State v. Harding, 141 Ariz. 492, 687 P. 2d 1247 *179(1984) (defendant killed victim); State v. Libberton, 141 Ariz. 132, 685 P. 2d 1284 (1984) (defendant killed victim); State v. Jordan, 137 Ariz. 504, 672 P. 2d 169 (1983) (defendant killed and intended to kill); State v. Smith, 138 Ariz. 79, 673 P. 2d 17 (1983) (defendant killed and intended to kill), cert. denied, 465 U. S. 1074 (1984); State v. Richmond, 136 Ariz. 312, 666 P. 2d 57 (defendant intended to kill, participated in assault that led to death), cert. denied, 464 U. S. 986 (1983); State v. McDaniel, 136 Ariz. 188, 665 P. 2d 70 (1983) (defendant killed victim); State v. Gillies, 135 Ariz. 500, 662 P. 2d 1007 (1983) (defendant took an active and deliberate part in the killing). Although the Court suggests otherwise, ante, at 155-156, n. 11, in none of these cases does the Arizona Supreme Court’s finding of intent appear to rest, as it did here, on a finding that a killing was merely foreseeable.
The utilitarian logic of deterrence can also justify unjust punishments that are more commonly dispensed. See Fletcher, Rethinking Criminal Law, at 415 (“Judges in traffic courts are readily tempted by the philosophy that regardless of whether the particular suspect has committed the violation, a punitive fine will make him drive more carefully in the future”).
A sophisticated utilitarian theory of deterrence might propose some limiting principles, e. g., “no punishment must cause more misery than the offense unchecked.” H. Hart, Punishment and Responsibility 76 (1968). But as Hart points out, this and other principles “do not seem to account for the character of the normal unwillingness to ‘punish’ those who have not broken the law at all, nor for the moral objection to strict liability which permits the punishment of those who act without mens rea.” Ibid. In Hart’s view, “civilized moral thought” would limit the utilitarian theories of punishment “by the demand that punishment should not be applied to the innocent,” and by limiting “punishments in order to maintain a scale for different offenses which reflects, albeit very roughly, the distinction felt between the moral gravity of these offenses. Thus we make some approximation to the ideal of justice of treating morally like cases alike and morally different ones differently.” Id., at 80. It is worth noting that both of the limits Hart identifies have been given vitality in the Court’s proportionality jurisprudence. E. g., Robinson v. California, 370 U. S. *181660, 667 (1962) (“Even one day in prison would be a cruel and unusual punishment for the ‘crime’ of having a common cold”); Enmund v. Florida, 458 U. S., at 801 (Enmund’s “punishment must be tailored to his personal responsibility and moral guilt”).
Such punishment might also be defended on the utilitarian ground that it was necessary to satisfy the community’s thirst for retribution and thereby keep the peace. Such grounds can be used to justify the punishment even of innocent people when the guilty have not been found and the mob threatens new violence. It is thus clear that “channeling” retributive instincts requires the State to do more than simply replicate the punishment that private vengeance would exact. To do less is simply to socialize vigilantism. As Justice MARSHALL has stated: “[T]he Eighth Amendment is our insulation from our baser selves. The ‘cruel and unusual’ language limits the avenues through which vengeance can be channeled. Were this not so, the language would be empty and a return to the rack and other tortures would be possible in a given case.” Furman v. Georgia, 408 U. S. 238, 345 (1972) (concurring opinion). See also Gregg v. Georgia, 428 U. S. 153, 237-241 (1976) (Marshall, J., dissenting) (death penalty unnecessary to further legitimate retributive goals).
The prophets warned Israel that theirs was “a jealous God, visiting the iniquity of the fathers upon the children unto the third and fourth generation of them that hate [Him].” Exodus, 20:5 (King James version). See, e. g., Horace, Odes III, 6:1 (C. Bennett trans. 1939) (“Thy fathers’ sins, 0 Roman, thou, though guiltless, shall expiate”); W. Shakespeare, The Merchant of Venice, Act III, scene 5, line 1 (“Yes, truly, for look you, the sins of the father are to be laid upon the children”); H. Ibsen, Ghosts (1881).