Opinion by Judge REINHARDT; Dissent by Judge THOMPSON.
REINHARDT, Circuit Judge:Following a single trial before “dual juries,” petitioners Joe Leonard Lambright and Robert Douglas Smith were convicted of first degree murder, kidnapping and sexual assault, and were sentenced to death. Because Arizona law did not authorize the use of dual juries at the time of petitioners’ trial, the Arizona Supreme Court concluded that the trial judge in this case had conducted an inappropriate and unauthorized experiment with petitioners’ trial. It nonetheless upheld the convictions. We hold that the Due Process Clause of the Fourteenth Amendment does not permit a trial judge in a capital case to engage in unauthorized experiments with a central element of the historic trial structure. Moreover, we also hold that, by failing to follow its own procedural rules, Arizona violated a state created liberty interest, and thereby deprived petitioners of their- federal due process rights. For these reasons, we reverse their convictions and remand for issuance of a writ of habeas corpus.
I.
Facts and Procedural History
Lambright and Smith were convicted of the March 11, 1980 kidnapping, sexual assault and brutal murder of Sandra Owen. The crimes occurred in the course of a cross-country trip taken by Lambright, Lam-bright’s girlfriend Kathy Foreman, and Smith. There is little doubt that both defendants committed serious criminal offenses, but each denied to the police that he participated in the actual murder: Lambright told the police that Smith and Foreman were involved, while Smith said it was Foreman and Lambright. Foreman turned state’s evidence and testified that she asked the two defendants to let Owen go, but they refused and then proceeded to murder her.
Prior to trial, the state moved to consolidate Lambright and Smith’s trials. Both petitioners opposed consolidation, and filed motions to sever the case. The trial judge agreed that the cases should be severed, but decided that rather than holding separate trials, as required by Arizona Rule of Criminal Procedure 13.4, he would conduct a single trial before two juries in one courtroom.
Over the objections of both defendants, the court then commenced a “dual jury” trial. Two juries were selected from separate veni-res. When evidence admissible against both defendants was presented, both juries remained in the courtroom. One sat in the jury box, the other in a row of chairs on the opposite side of the courtroom. The juries switched places every other day. When evidence admissible as to only one defendant was offered, that defendant’s jury would remain in the courtroom while the other jury was excused. Separate opening and closing arguments were made, and the juries were instructed separately and deliberated separately.
On March 30, 1982 the dual juries found Lambright and Smith guilty of first degree murder, sexual assault and kidnapping. Two months later, Judge Michael Brown sentenced both petitioners to death.
On direct appeal to the Arizona Supreme Court, both Lambright and Smith argued that there was no basis in state law for the use of dual juries. Both petitioners also alleged that the use of dual juries deprived them of due process under the Fourteenth Amendment to the United States Constitution. The Arizona Supreme Court largely agreed with petitioners’ assertions, holding that the use of dual juries was in fact unauthorized by state law. See State v. Lambright, 138 Ariz. 63, 673 P.2d 1, 7 (Ariz.1983), cert. denied, 469 U.S. 892, 105 S.Ct. 267, 83 L.Ed.2d 203 (1984); State v. Smith, 138 Ariz. 79, 673 P.2d 17, 19 (1983), cert. denied, 465 U.S. 1074, 104 S.Ct. 1429, 79 L.Ed.2d 753 (1984). The court, however, engaged in a harmless error review, and finding that no prejudice resulted from the use of the “experiment[al]” proceeding, affirmed the convictions and sentences. See State v. Lambright, 673 P.2d at 7; State v. Smith, 673 P.2d at 19.
*480Both petitioners sought post-conviction review in the Arizona state courts pursuant to Arizona Rule of Criminal Procedure 32. Lambright filed his first Rule 32 petition in 1984.1 The state court denied relief, and set the execution date. Lambright then filed his original habeas petition in the United States District Court for the District of Arizona, in which he argued, inter alia, that the use of dual juries violated Arizona law and his Fourteenth Amendment due process rights. The district court stayed the execution, but dismissed Lambright’s petition without prejudice because, the court held, it contained unexhausted claims. In 1991, Lambright filed his second Rule 32 petition in Arizona state court. The state court again denied the petition. In 1995, Lambright filed a third Rule 32 petition in state court, which was also denied.
In 1995, Lambright filed an amended petition for writ of habeas corpus in the federal district court. Before the district court, Lambright reiterated his arguments that the use of dual juries violated the Fourteenth Amendment. He also argued that the use of dual juries was unauthorized by Arizona law at the time of his trial, and that by depriving him of the type of jury trial to which he was entitled under Arizona law, the trial court deprived him of his Fourteenth Amendment Due Process Rights. The district court denied the habeas petition. See Lambright v. Lewis, 932 F.Supp. 1547 (D.Ariz.1996).
Smith filed his first Rule 32 petition in 1984, which the state court denied. In 1987, Smith filed a habeas petition in the United States District Court for the District of Arizona, asserting, inter alia, that the use of dual juries was unauthorized under Arizona law and that their use deprived him of his Fourteenth Amendment due process rights. The district court stayed the execution and dismissed the petition without prejudice for failure to exhaust state court remedies. Smith then filed a second Rule 32 petition in 1989, which was denied by the state court. Smith’s third and final Rule 32 petition was also denied.2
In 1994, Smith filed his amended petition for writ of habeas corpus in the district court. In this petition Smith argued that the use of dual juries violated the Arizona Rules of Criminal Procedure and violated his due process rights under the Fourteenth Amendment. The district court denied the amended habeas petition. Smith v. Lewis, No. CIV 87-234 TUC RMB, Amended Memorandum Opinion and Order (D.Ariz.l996)[Amended Order].
II.
The Unauthorized Use of Dual Juries
A. Procedural Bar
It is well settled that before a claim may be reviewed by a federal habeas court, the factual and legal basis for that claim must be presented in state court so that the state has a reasonable opportunity to address the claim and correct any violation of federal rights. See Duncan v. Henry, 513 U.S. 364, 365, 115 S.Ct. 887, 130 L.Ed.2d 865 (1995). Where the claim involves a violation of state law, petitioners must specifically allege not only the state law violation, but the federal constitutional right they are asserting as well. See id.; see also Anderson v. Harless, 459 U.S. 4, 103 S.Ct. 276, 74 L.Ed.2d 3 (1982).
The district court did not find Smith’s dual jury claim proeedurally barred. See Smith v. Lewis, Amended Order at 7. This holding was correct. In his direct appeal to *481the Arizona Supreme Court, Smith argued that the use of dual juries was unauthorized by state law and that their use deprived him of his Fourteenth Amendment due process rights. Accordingly, Smith adequately presented this claim in state court.
Lambright presented identical arguments in his direct appeal. Like Smith, Lambright argued, clearly and specifically, that the use of dual juries was unauthorized under state law. He also clearly alleged a violation of his federal due process rights under the Fourteenth Amendment. He, therefore, specifically alleged both the state law violation and the federal constitutional right at issue, and the Arizona Supreme Court accordingly was afforded the opportunity to address the claim.
Despite the fact that Lambright and Smith made precisely the same dual jury arguments in their direct appeals, the district court found Lambright’s claim procedurally barred. Curiously, the district court held that Lambright failed to raise the claim in state court. See Lambright v. Lewis, 932 F.Supp. at 1568.3 The district court based its holding on the theory that the “factual and legal basis for a claim must be presented to the state court so that the state has a reasonable opportunity to address the claim_” Id. at 1568. Contrary to the holding of the district court, and contrary to what the state has argued before this court, however, the record makes it clear that Lam-bright adequately presented the claim to the state court in his direct appeal. The district court’s conclusion that Lambright’s claim is procedurally barred was therefore erroneous.
As neither petitioner is procedurally barred from bringing the dual jury claims, we proceed to the merits.4
B. Merits
On direct appeal to the Arizona Supreme Court, both petitioners claimed that there was no basis in state law for a dual jury trial, and that the use of two juries deprived them of due process of law under the Fourteenth Amendment to the United States Constitution. The Arizona Supreme Court largely agreed. It held that the Arizona Rules of Criminal Procedure did not authorize a trial court to conduct a trial with dual juries, writing that “[i]t is clear that the ‘experiment’ conducted in the instant case was unauthorized— ” State v. Lambright, 138 Ariz. 63, 673 P.2d 1, 7 (Ariz.1983). The Arizona Supreme Court also noted, rightly of course, that “death penalty cases are inappropriate vehicles for experimentation with new procedures.” Id. at 8.
Although the Arizona Supreme Court held that Lambright and Smith had been convicted in a proceeding that constituted an inappropriate and unauthorized experiment, it went on to consider whether petitioners had been prejudiced by the use of dual juries. Because the petitioners were not able to point to a “specific error occurring at trial,” the court affirmed the convictions. See id.
1. Direct Federal Due Process Violation
a. Merits
It should go without saying that a trial judge may not, consistent with the Due Process Clause of the Fourteenth Amendment, reinvent the basic procedures or mechanism by which a capital defendant is tried. Due process, which demands procedural regularity, simply does not permit a judge to conduct unauthorized experiments with the fundamental structure of capital trials. Yet, as the Arizona Supreme Court held, this is exactly what happened at petitioners’ trial. The Arizona trial judge decided, unilaterally, to conduct an unauthorized experiment by employing a dual jury. Because the Due *482Process Clause of the Fourteenth Amendment guarantees criminal defendants something more than an unauthorized experimental trial, we reverse the convictions.
Although federal due process protections extend to all state criminal trials, the Supreme Court has demanded an even higher degree of procedural regularity and reliability in capital cases. The heightened requirement of procedural regularity and reliability derives from the longstanding recognition that:
death is a different kind of punishment from any other which may be imposed in this country.... From the point of view of the defendant, it is different in both its severity and its finality. From the point of view of society, the action in taking the life of one of its citizens also differs dramatically from any other legitimate state action. It is of vital importance to the defendant and to the community that any decision to impose the death sentence be, and appear to be, based on reason rather than caprice or emotion.
Gardner v. Florida, 430 U.S. 349, 357-58, 97 S.Ct. 1197, 51 L.Ed.2d 393; see also Ford v. Wainwright, 477 U.S. 399, 411, 106 S.Ct. 2595, 91 L.Ed.2d 335 (1986) (plurality opinion) (“In capital proceedings generally, this Court has demanded that factfinding procedures aspire to a heightened standard of reliability.... This especial concern is a natural consequence of the knowledge that execution is the most irremediable and unfathomable of penalties....”)
Indeed, in numerous opinions, the Supreme Court has held that the death penalty may be imposed only following rigidly regular and reliable sentencing procedures. The Court has therefore dictated the discretion that sentencing courts and juries must have. See, e.g., Lockett v. Ohio, 438 U.S. 586, 98 S.Ct. 2954, 57 L.Ed.2d 973 (1978); Godfrey v. Georgia, 446 U.S. 420, 100 S.Ct. 1759, 64 L.Ed.2d 398 (1980). Courts are required to scrutinize the wording of jury instructions. See, e.g., Caldwell v. Mississippi 472 U.S. 320, 105 S.Ct. 2633, 86 L.Ed.2d 231 (1985); Mills v. Maryland, 486 U.S. 367, 108 S.Ct. 1860, 100 L.Ed.2d 384 (1988). They must, moreover, set strict guidelines for the type of evidence that may be admitted, must be admitted, and may not be admitted. See, e.g., Skipper v. South Carolina, 476 U.S. 1, 106 S.Ct. 1669, 90 L.Ed.2d 1 (1986); Booth v. Maryland, 482 U.S. 496, 107 S.Ct. 2529, 96 L.Ed.2d 440 (1987).
In Beck v. Alabama, the Supreme Court made explicit the obvious point that similar concerns and heightened protections extend to the guilt phase of capital trials. The Beck Court held that the federal Due Process Clause requires lesser included offense instructions in all capital trials, although it reserved the question of whether due process requires such instructions in noncapital cases. The Court concluded that “[t]o insure that the death penalty is indeed imposed on the basis of ‘reason rather than caprice or emotion,’ we have invalidated procedural rules that tended to diminish the reliability of the sentencing determination. That same reasoning must apply to rules that diminish the reliability of the guilt determination.” Beck, 447 U.S. at 637-38, 100 S.Ct. 2382 (internal citations omitted).
This court has also emphasized the higher level of procedural reliability demanded of capital trials. In Beam v. Paskett, 3 F.3d 1301 (1993), a case in which we examined the authorized use of dual juries, we stressed the heightened procedural requirements of capital cases. We wrote that cases that result in the imposition of the death penalty require “exacting constitutional scrutiny.” Id. at 1303-04. We also held that “the level of constitutional scrutiny given to the state’s use of a dual jury procedure is determined by the nature of the punishment the defendant will actually suffer.... Whatever additional constitutional constraints exist on the use of dual juries in capital trials would be a consequence of the greater reliability demanded of verdicts upon which a sentence of death is based_” Id. at 1304. In expressing its own disapproval of the trial judge’s use of dual juries in Lambright’s and Smith’s trial, the Arizona Supreme Court wrote that “death penalty cases are inappropriate vehicles for experimentation with new procedures.” State v. Lambright, 138 Ariz. 63, 673 P.2d 1, 8 (Ariz.1983).
*483States may, of course, “experiment” with new procedures designed to improve the trial process. Indeed, the ability of states to serve as “laboratories” has long been heralded as one of the “happy incidents” of our federal system. See New State Ice Co. v. Liebmann, 285 U.S. 262, 280, 52 S.Ct. 371, 76 L.Ed. 747 (1932)(Brandeis, J., dissenting); see also Arizona v. Evans, 514 U.S. 1, 8, 115 S.Ct. 1185, 131 L.Ed.2d 34 (1995) (citing New State Ice Co., 285 U.S. at 280).
Perhaps the most well-known endorsement of state court “experimentation” with trial procedures came in Chandler v. Florida, 449 U.S. 560, 101 S.Ct. 802, 66 L.Ed.2d 740 (1981), where the Court held that a state could provide for electronic media coverage of criminal trials. While the Court approved of Florida’s “experimentation” with the use of cameras in the criminal courtroom, before granting its approval the Court explained in great detail the procedures by which Florida had authorized the experiment, Id. 101 S.Ct. at 803-05. Several television stations petitioned the Supreme Court of Florida to change Florida’s Canon of Judicial Ethics 3A(7), which proscribed television coverage of courtroom proceedings. Only after the Florida Supreme Court had “invited presentations in the nature of a rulemaking proceeding” did it allow the televising of one civil and one criminal trial, under specific guidelines. The state supreme court then:
received and reviewed briefs, reports, letters of comment, and studies. It conducted its own survey of attorneys, witnesses, jurors, and court personnel- A separate survey was taken of judges by the Florida Conference of Circuit Judges. The court also studied the experience of 6 States that had, by 1979, adopted rules relating to electronic coverage of trials, as well as that of 10 other States that, like Florida, were experimenting with such coverage.
Id. 101 S.Ct. at 805. Following this extensive period of review the Florida Supreme Court concluded that “on balance there [was] more to be gained than lost by permitting electronic media coverage of judicial proceedings subject to standards for such coverage.” Id. (quoting In re Petition of Post-Newsweek Stations, Florida, Inc., 370 So.2d 764, 780 (1979)).
When states carefiilly consider the impact of new techniques and procedures on a criminal trial, they may, within limits, authorize experimentation with those techniques. Because Florida engaged in such reasoned consideration of the impact that television would have on trials, the United States Supreme Court approved the experiment. This obviously does not imply, however, that individual state trial judges have free reign to experiment with the central elements of the historic trial structure when the state itself has not considered whether to authorize its judges to do so. To allow this type of unauthorized experimentation in a capital case would be to allow individual judges to determine the quality of process each capital defendant would receive. Our constitution guarantees “due process” to prevent exactly this.
Before carrying out an experiment with the fundamental elements of the criminal process, a state must proceed cautiously. It must give due consideration to the potential impact of the procedural innovation. In 1982, Arizona did have a set of rules for the authorization of new criminal procedures. Trial courts had the authority to propose changes, but could not implement them without first securing the approval of the Arizona Supreme Court. As the court stated in this very case:
Nothing we say here should discourage courts, through the adoption of local rules, to carry out experiments which may improve the judicial process. Indeed, these efforts should be encouraged. But local rules must first be approved by this court.
Lambright, 673 P.2d at 7 (emphasis added).5
The trial judge in this case did not take the steps Arizona required for the authorization of experimental trial procedures. Instead, *484he simply decided unilaterally that two defendants would be tried, in a capital case, in an experimental proceeding that he, himself, designed.6 Nothing Justice Brandéis ever said about the virtue of states as laboratories comes close to sanctioning this type of unguided experiment. In a capital case where procedural protections are heightened and courts engage in “exacting constitutional scrutiny,” Beam, 3 F.3d at 1303-04, a trial judge simply may not design his own criminal procedures. Where he does so, the Due Process Clause of the Fourteenth Amendment is violated.
Our decision is not affected by the fact that Lambright was overruled nine years after it was decided. In. Hedlund v. Sheldon, 173 Ariz. 143, 840 P.2d 1008 (Ariz.1992), the Arizona Supreme Court held that it is within the discretion of a trial judge to use a dual jury in a trial of two co-defendants. See id. 840 P.2d at 1011. The court concluded that the decision to empanel such a jury did not amount to a “local rule,” and therefore could be made by the judge alone. See id. As the Hedlund court explained, however, its decision to overrule Lambright was based on new “developments in court procedures and changing circumstances,” including a “recent” recognition on the part of the Arizona courts of the importance of “according flexibility and discretion to judges.” Id. at 1009. It was only “in light of these circumstances” that the Hedlund court found it “appropriate to reassess LambrighV’ Id. at 1010. The Hedlund court’s reliance- on changed circumstances makes it clear that its 1992 decision does not apply to trials conducted a decade earlier.7
b. Type of Error
Having determined that the unauthorized experimentation conducted by the trial judge in this case constitutes a due process violation, our next task is to determine what type of error — trial or structural- — results from the use of an unauthorized, experimental jury system.8 Structural er*485rors are those that affect “the framework within which the trial proceeds.” Arizona v. Fulminante, 499 U.S. 279, 309, 111 S.Ct. 1246, 113 L.Ed.2d 302 (1991). They are “structural defects in the constitution of the trial mechanism, which defy analysis by ‘harmless error’ standards,” id., and “affect the trial from beginning to end.” Rice v. Wood, 77 F.3d 1138, 1141 (9th Cir.1996). Trial errors, on the other hand, are generally errors that occur “during the presentation of the ease to the jury, and that may therefore be quantitatively assessed in the context of other evidence presented.” Id. at 307-08, 111 S.Ct. 1246. Trial errors, unlike structural errors, are subject to harmless error analysis. We have held that “[t]he purpose of harmless-error analysis is to avoid setting aside convictions for small errors or defects that have little, if any, likelihood of having changed the result of the trial.” United States v. Annigoni, 96 F.3d 1132, 1143 (9th Cir.1996)(internal citations omitted).
Our determination of what type of error occurred here, where the trial judge impermissibly experimented with the jury structure, is aided by the Supreme Court’s decision in Sullivan v. Louisiana, 508 U.S. 275, 113 S.Ct. 2078, 124 L.Ed.2d 182 (1993). In Sullivan, the Court held that a jury instruction that erroneously defined the reasonable doubt standard effectively deprived the defendant of his right to a jury trial. The Court first distinguished “structural defects in the constitution of the trial mechanism,” from “errors which occur during the presentation of the case to the jury.” Id. at 282, 113 S.Ct. 2078 (internal citations omitted). It then held that an erroneous reasonable doubt instruction constitutes structural error, writing:
[T]he jury guarantee [is] a ‘basic protection’ whose precise effects are unmeasurable, but without which a criminal trial cannot reliably serve its function. The right to trial by jury reflects, we have said, a ‘profound judgment about the way in which law should be enforced and justice administered.’ The deprivation of that right, with consequences that are necessarily unquantifiable and indeterminate, unquestionably qualifies as ‘structural error.’
Sullivan, 508 U.S. at 282, 113 S.Ct. 2078 (internal citations omitted).
It is, indeed, hard to imagine an error more starkly structural than an error implicating the very design of the jury mechanism. When, as here, a judge impanels an unauthorized and experimental jury system, he commits an error of this type.9 The consequences of such an error are, as the Sullivan Court wrote, “necessarily unquanti*486fiable and indeterminate.” Id. Such an error, moreover, quite obviously amounts to an error in the “constitution of the trial mechanism,” that affects “the framework within which the trial proceeds.” Fulminante, 499 U.S. at 309, 111 S.Ct. 1246. We hold, therefore, that an error in the fundamental design of the jury mechanism “unquestionably qualifies as structural error.” Id.10
It is well settled that structural errors are not subject to harmless error review. See, e.g. Brecht v. Abrahamson, 507 U.S. 619, 629, 113 S.Ct. 1710, 123 L.Ed.2d 353 (1993); Sullivan, 508 U.S. at 280, 113 S.Ct. 2078; Arizona v. Fulminante, 499 U.S. 279, 309, 111 S.Ct. 1246, 113 L.Ed.2d 302 (1991). This is true even on federal habeas review of state convictions. See Crandell v. Bunnell, 144 F.3d 1213, 1216 (9th Cir.1998); Bland v. California Dep’t of Corrections, 20 F.3d 1469, 1477 (9th Cir.1994) (citing Brecht, 507 U.S. at 629-30, 113 S.Ct. 1710). Where a structural error occurs at trial, an appeals court does not conduct a harmless error review. Nor do we look for a specific showing of prejudice. As the Supreme Court has held, “[t]he existence of such defects ... requires automatic reversal of the conviction because they infect the entire trial process.” Brecht, 507 U.S. at 629-30, 113 S.Ct. 1710 (citing Fulminante, 499 U.S. at 309-10, 111 S.Ct. 1246).
Because the use of dual juries at Lam-bright and Smith’s trial constituted structural error, we reverse petitioners’ convictions.
2. Violation of a State Created Liberty Interest
Even if the Due Process Clause permitted the type of inappropriate and unauthorized experimentation conducted by the trial judge in this case, the use of dual juries violated a state created liberty interest and therefore deprived Lambright and Smith of their federal due process rights.
Where a defendant is deprived of a state statutory right, such a deprivation *487may implicate the federal Due Process Clause. States may create “liberty interests” that are protected by the Fourteenth Amendment. See Hicks v. Oklahoma, 447 U.S. 343, 346, 100 S.Ct. 2227, 65 L.Ed.2d 175 (1980). As this court has held on more than one occasion, “the failure of a state to abide by its own statutory commands may implicate a liberty interest protected by the Fourteenth Amendment against arbitrary deprivation by a state.” Fetterly v. Paskett, 997 F.2d 1295, 1300 (9th Cir.1993); see also e.g., Ballard v. Estelle, 937 F.2d 453, 456 (9th Cir.1991).
In Hicks v. Oklahoma, 447 U.S. 343, 344-45, 100 S.Ct. 2227, 65 L.Ed.2d 175 (1980), the jury was instructed that if it finds Hicks guilty it “shall assess [the] punishment at forty (40) years imprisonment.” An Oklahoma statute in effect at the time of Hicks’ trial, however, required that sentences be fixed by the jury. The Court rejected Oklahoma’s argument that the denial of this state procedural right was “of exclusively state concern”:
Where ... a State has provided for the imposition of criminal punishment in the discretion of the trial jury, it is not correct to say that the defendant’s interest in that discretion is merely a matter of state procedural law. The defendant in such a case has a substantial and legitimate expectation that he will be deprived of his liberty only to the extent determined by the jury in the exercise of its statutory discretion, ... and that liberty interest is one that the Fourteenth Amendment preserves against arbitrary deprivation by the state.
Id. at 346, 100 S.Ct. 2227. Because the jury instruction prevented the jury from exercising the discretion afforded by state statute, the Court held that Hicks was denied the jury sentence to which he was entitled under state law. As a result, the Court held, Oklahoma had denied Smith his federal due process rights.
Following the logic of Hicks, this court has similarly held that deprivations of state statutory rights can implicate the federal Due Process Clause. In Ballard, the petitioner claimed that his sentence violated California sentencing law because the sentencing court had used a broader definition of “use of a firearm” than the California sentencing laws allowed. Id. at 455-56. After noting that “[s]tate laws may give rise to liberty interests protected by the Fourteenth Amendment,” we held that Ballard’s sentencing claim “set[ ] forth a cognizable federal habe-as corpus claim based on the due process clause of the Fourteenth Amendment.” Id. at 456.
In Fetterly v. Paskett, moreover, we held that by mandating an aggravating/mitigating circumstance balancing regime for the imposition of the death penalty, Idaho had created a liberty interest. See Fetterly, 997 F.2d at 1300. Because there was “dissonance” between the Idaho sentencing law and the manner in which Fetterly was actually sentenced, id. at 1299, we held that Fetterl^s sentence raised federal due process concerns, and his claim was therefore cognizable on federal habeas review. See id. at 1301.
The case before us bears marked similarities to Fetterly. While the dissonance there was between the requirement of the sentencing law and the procedure actually followed, here the dissonance was between the requirements of the Arizona Rule — which required co-deferidants to be tried only in severed jury trials, or in a single trial before one jury — and the jury mechanism employed by the district court. See Ariz.R.Crim.P. 13.4; State v. Lambright, 673 P.2d at 7. In both cases, the dissonance was caused by a failure to employ a state mandated regime and in both cases the result was the violation of a state created liberty interest.
Of course, all violations of state law do not implicate federal due process concerns. It is only when a state statute creates a “liberty interest” that the violation of state law raises federal constitutional concerns cognizable on federal habeas review. In previous cases, we have delineated the attributes a state rule must possess in order to create a liberty interest protected by the Due Process Clause. First, the law must contain “substantive predicates” that circumscribe official decisionmaking. See Bonin v. Calderon, 59 F.3d 815, 842 (9th Cir.1995). Second, the law must contain “explicitly mandatory language.” See id. Finally, the state law “must *488provide more than merely procedure; it must ■ protect some substantive end.” Id. (quoting Dix v. County of Shasta, 963 F.2d 1296, 1299 (9th Cir.1992)).
Article 2 of the Arizona Constitution protects the right of criminal defendants to trial by jury. Article 2, § 23 of the constitution provides that “[t]he right of trial by jury shall remain inviolate.” In 1982, the Arizona Rules of Criminal Procedure, as interpreted by the Arizona Supreme Court, dictated that in a trial of co-defendants, the right to a trial by jury could be realized in one of two ways: first, by severed, individual jury trials; or, second, by a single trial before one jury. See State v. Lambright, 673 P.2d at 7; Ariz.R.Crim.P. 13.4 Rule 13.4 thereby established a substantive predicate circumscribing government action: it dictated the only two ways in which a trial judge could structure a trial of co-defendants. The language of the Rule, moreover, was explicit and mandatory: it allowed two, and only two, alternatives for effecting the right to trial by jury in a trial of co-defendants. In addition, as the Arizona Supreme Court pointed out in Lambright, 673 P.2d at 7, according to Arizona Rule of Criminal Procedure 36, changes or modifications to the rules are prohibited unless first approved by that court. Finally, Rule 13.4 clearly protected “some substantive end”: it protected the integrity of the fact-finding process and the right to a fair jury trial. We conclude, therefore, that Arizona Rule of Criminal Procedure 13.4 created a state liberty interest. It mandated that Arizona’s guarantee of a jury trial could be realized in a trial of codefendants only through severed, individual trials, or by a single trial before one jury.
The state trial court did not provide petitioners either of these alternatives. Rather, it provided them with an unauthorized, experimental proceeding. The court thereby deprived petitioners of the jury trial to which they were entitled under Arizona law. Such a deprivation amounts to a violation of a state created liberty interest, and thus to a violation of the petitioners’ federal due process rights.11 While the authorized use of dual juries may not result in a direct federal constitutional violation,12 where the use of such juries deprives a defendant of a state created liberty interest, such use constitutes error cognizable on federal habeas review.13
This conclusion, however, does not end our inquiry. Again, we must also determine what type of error — trial or structural — results from the use of dual juries when such use deprives a defendant of the jury process to which he is entitled by a state created liberty interest. We have already answered this question in Section IB1. There are few errors more patently “structural” than the deprivation of the right to the type of jury process guaranteed by law. The Supreme Court has made this clear. See Sullivan, 508 U.S. at 282, 113 S.Ct. 2078.
Under Arizona law, as it existed in 1982, Lambright and Smith were not tried by the type of jury that state law guaranteed them. They were tried instead by a judge-created jury mechanism, “unauthorized” and “experiment[al].” They were therefore deprived of their right to the jury trial guaranteed them by Arizona law. The consequences of the *489deprivation of this right are “unquantifiable and indeterminate.” See id. The error is “unquestionably” structural. See id.
We repeat: structural errors are not subject to harmless error review, see, e.g. Brecht v. Abrahamson, 507 U.S. 619, 629, 113 S.Ct. 1710, 123 L.Ed.2d 353 (1993); Sullivan, 508 U.S. at 280, 113 S.Ct. 2078; Arizona v. Fulminante, 499 U.S. 279, 309, 111 S.Ct. 1246, 113 L.Ed.2d 302 (1991); this is true even on federal habeas review, see Crandell v. Bunnell, 144 F.3d 1213, 1216 (9th Cir.1998); Bland v. California Dep’t of Corrections, 20 F.3d 1469, 1477 (9th Cir.1994) (citing Brecht, 507 U.S. at 629-30, 113 S.Ct. 1710). The existence of such errors requires automatic reversal of the petitioners’ convictions. Brecht, 507 U.S. at 629-30, 113 S.Ct. 1710 (citing Fulminante, 499 U.S. at 309-10, 111 S.Ct. 1246).14 Accordingly, reversal is required on this ground as well.
III.
Conclusion
The constitutional guarantees of due process extend to all defendants “regardless of the heinousness of the crime [and] the apparent guilt of the offender.” Irvin v. Dowd, 366 U.S. 717, 722, 81 S.Ct. 1639, 6 L.Ed.2d 751 (1961). The petitioners here are entitled to new trials. We therefore REVERSE the district court’s denial of petitioners’ habeas petitions, and REMAND the ease to the district court with instructions to grant the writ and to direct the state to release the petitioners unless the state retries them within a reasonable time.
REVERSED AND REMANDED.
. Beginning with his first Rule 32 petition, Lam-bright has presented claims of ineffective assistance of counsel in each of his state and federal post-conviction proceedings. Specifically, he alleges that he received ineffective assistance when his counsel failed to present mitigating psychiatric testimony during the sentencing phase of Lambright’s trial. Although we do not reach the issue here because we reverse Lambright’s conviction, we do note that Lambright's claims appear to have substantial merit.
. In his third Rule 32 petition, and again in a supplemental habeas corpus petition submitted to the district court, Smith raised serious allegations of ineffective assistance of counsel at sentencing and offered colorable justifications for his failure to raise them previously. Because we reverse Smith’s conviction, we do not reach those claims here. We note, nevertheless, that the claims, like Lambright’s, appear to have substantial merit.
. The district court found the claims technically exhausted and procedurally defaulted because, it held, if Lambright did return to the state court to raise the claim the state court would determine that he had procedurally defaulted it by not raising it on direct appeal. See id. at 1568-69.
. On appeal to this court, petitioners again allege that the use of dual juries violates their due process rights under the Fourteenth Amendment. Their claim is broader than required for resolution of the issue. We do not determine whether the use of dual juries in a capital punishment case directly violates the constitution, a question we have explicitly left open. See Beam v. Paskett, 3 F.3d 1301 (1993).
. The Arizona Supreme Court had "exclusive power to make rules relative to all procedural matters in any court.” That rule-making power could "not be supplemented by a Superior Court." Id. (quoting Hare v. Pima County Superior Ct., 133 Ariz. 540, 652 P.2d 1387, 1389 (1982)).
. For this reason the Arizona Supreme Court concluded that "[i]t is clear that the 'experiment’ conducted in the instant case was unauthorized. ...” 673 P.2d at 7.
. It is also worth noting that in State v. Slemmer, 170 Ariz. 174, 823 P.2d 43 (Ariz. 1991), the Arizona Supreme Court adopted the following retro-activity standard: "decisions overruling precedent and establishing a new rule are 'almost automatically nonretroactive’ to cases that are final and are before the court only on collateral attack." Id. at 47 (quoting Allen v. Hardy, 478 U.S. 255, 258, 106 S.Ct. 2878, 92 L.Ed.2d 199 (1986)).
. Our dissenting colleague argues that "[t]he use of dual juries does not rise to the level of structural error.” (Dissent at 490). This case, however, is not simply about "the use of dual juries” standing alone. As we make clear throughout our opinion, we are called on to assess the inappropriate, unauthorized, and experimental use of such juries in a capital trial. That is, the case calls on us to determine what type of error occurs when a trial judge conducts an inappropriate, unauthorized experiment with the fundamental elements of the criminal process — in disregard of the rules established by tire state for the authorization of experimental trial procedures— and does so in a capital trial. The case does not call on us to determine what type of error occurs from the properly authorized use of dual juries in non-capital proceedings. This latter question has already been resolved, as the dissent correctly points out. See Beam v. Paskett, 3 F.3d 1301 (9th Cir.1993). But so has the fact that the use of such a process in capital cases must be viewed differently, even when the procedure has been properly authorized — as it was not here.
None of the cases cited by Judge Thompson involves the unauthorized and experimental use of dual juries in a capital proceeding, and therefore none presents the question raised by the case at bench. In none of those cases did a trial judge in a capital proceeding unilaterally implement an experimental proceeding that he himself designed, in contravention of the steps required by law for the authorization of experimental trial procedures. In short, none of the cases cited by the dissent presents anything like the egregious error before us; none disregards the fundamental trial structure established by applicable law.
Judge Thompson attempts to avoid the critical differences between the cases he cites and tire case at bench by writing that “[djistinguishing between the use of dual juries in a non-capital case and a capital case creates a distinction without a difference.” (Dissent at 492, citing Beam, 3 F.3d at 1304.) Of course, as we demonstrate above, the difference between the procedural regularity and reliability demanded of capital and non-capital trials is anything but a "distinction without a difference.” Indeed, the Supreme Court’s death penalty jurisprudence is often captured by the phrase "death is different.” See infra Opinion at 481-482. Moreover, the Beam opinion itself made explicit this court’s long-established understanding that capital trials demand heightened procedural regularity, writing of the "greater reliability demanded of ver*485dicts upon which a sentence of death is based.” 3 F.3d at 1304. That case also relied explicitly on the difference between capital and non-capital trials, reaching only the question of whether the use of dual juries in non-capital trials violated the constitution absent a showing of prejudice, and noting that "we have never considered the question in connection with a capital case." Id. at 1303.
. The dissent mischaracterizes our conclusion, writing that "Although the use of dual juries necessarily affects the mechanism of a trial, this circumstance standing alone does not equate to structural error.... The majority simply assumes ... that employing dual juries in a joint-defendant trial constitutes a defect in the constitution of the trial mechanism.” Of course, we assume nothing of the sort, nor do we conclude anything about the use of dual juries "standing alone.” Again, the case at bench is not about employing dual juries simpliciter. Rather, it is about a trial judge’s experimental use of dual juries, in contravention of the state law governing the use of experimental trial techniques, in a capital case. Our conclusion that a structural error has occurred is not based simply on the fact that dual juries were "employed” in the state court trial. On the contrary, we conclude only that when a trial judge conducts an "inappropriate,” "unauthorized,” "experiment" — in the words of the Arizona Supreme Court — with the fundamental elements of a capital trial, the “constitution of the trial mechanism” is implicated as is the “framework within which the trial proceeds.” Fulminante, 499 U.S. at 309, 111 S.Ct. 1246.
Finally, the dissent concludes that our "classification” is not supported by our “premise,” (Dissent at 492), only because it misconstrues our premise. To repeat: we do not presume, as the dissent would suggest, that the use of dual juries, alone, amounts to an error in the constitution of the trial mechanism. Rather, we conclude only that where the judge unilaterally, and in contravention of the procedures established by state law, structures a capital trial with an unauthorized and experimental jury, then there is an error in the "constitution of the trial mechanism” that affects “the framework within which the trial proceeds.” Id.
. The distinction between trial and structural errors is not always perfectly clear, and has been described as a "spectrum of constitutional errors,” see Brecht v. Abrahamson, 507 U.S. 619, 629, 113 S.Ct. 1710, 123 L.Ed.2d 353 (1993), rather than a "rigid dichotomy." United States v. Annigoni, 96 F.3d 1132, 1143 (9th Cir.1996). In Sullivan, for example, in addition to holding that the erroneous reasonable doubt instruction was structural error, the Court also concluded separately that such error is not amenable to harmless error analysis. The Court reasoned that harmless error analysis requires an appellate court to ask "not what effect the constitutional error might generally be expected to have upon a reasonable jury, but rather what effect it had upon the guilty verdict in the case at hand.” Id. at 279, 113 S.Ct. 2078. The Court concluded that “[t]he inquiry, in other words, is not whether, in a trial that occurred without the error, a guilty verdict would surely have been rendered, but whether the guilty verdict actually rendered in this trial was surely unattributable to the error.” Id. Where the appellate court cannot ascertain what effect the error had on the verdict rendered, but can only speculate about what a hypothetical jury might have done, the error is structural and therefore not amenable to harmless error review. See id. at 280, 113 S.Ct. 2078.
Because we conclude the error here is structural, it is not necessary for us also to determine whether the error is otherwise not amenable to harmless error analysis. Nevertheless, Sullivan 's reasoning clearly covers this case. Where, as here, the error occurs in the very design of the jury mechanism, it is, of course, impossible for an appellate court to know whether "the guilty verdict[s] actually rendered in this trial [were] surely unattributable to the error.” Sullivan, 508 U.S. at 279, 113 S.Ct. 2078. The body that tried Smith and Lambright was not the jury demanded by Arizona law, but was an "inappropriate” "unauthorized” "experiment." State v. Lambright, 138 Ariz. 63, 673 P.2d 1, 7-8 (Ariz.1983). As was the case in Sullivan, should we attempt harmless error review here, we would be forced to "speculate] about a hypothetical jury's action.” Sullivan, 508 U.S. at 280, 113 S.Ct. 2078.
In United States v. Annigoni, 96 F.3d 1132, 1144 (9th Cir.1996), this court determined that the error at issue was not amenable to harmless-error analysis, and thus required automatic reversal of the conviction, but did not reach the question of whether the error was appropriately classified as structural. See id. As we stated in that case:
In [the Fulminante Court’s] harmless-error analysis, the Court looked at the erroneously admitted evidence and compared it to other evidence the government had mustered against the defendant. It was able to evaluate the impact of the error because it was able to weigh the coerced confession relative to other known and tangible evidence.
Here, such weighing and quantifying is simply impossible because the error occurred in the design of the juiy mechanism.
While we hold that the error at issue in this case is structural error, it also appears abundantly clear that the error is not amenable to harmless error review.
. Sandin v. Conner, 515 U.S. 472, 482-84, 115 S.Ct. 2293, 132 L.Ed.2d 418 (1995), is not to the contrary. In Sandin, the Court held that a "negative implication from mandatory language in prisoner regulations " does not necessarily create a protected interest. Id. at 483, 115 S.Ct. 2293. This court has, quite appropriately, read Sandin as restricted to the prison context. See Picray v. Sealock, 138 F.3d 767 (9th Cir.1998); Carlo v. Chino, 105 F.3d 493, 498 (9th Cir.1997); Mitchell v. Dupnik, 75 F.3d 517, 524 (9th Cir.1996). While Chaney v. Stewart, 156 F.3d 921 (9th Cir.1998), a non-prison litigation case, cited Sandin, there was no issue in Chaney regarding a negative implication. Therefore, Chaney's reference to Sandin is simply dicta.
. See supra n. 4.
. It is of course true that federal habeas review is generally not available for errors of state law. See, e.g., Estelle v. McGuire, 502 U.S. 62, 71-72, 112 S.Ct. 475, 116 L.Ed.2d 385 (1991); Gerlaugh v. Stewart, 129 F.3d 1027, 1032 (9th Cir.1997)(citing Estelle, 502 U.S. at 67-68, 112 S.Ct. 475). Where the state creates a liberty interest, however, it is not correct to say that the question is "merely a matter of state procedural law." Hicks, 447 U.S. at 346, 100 S.Ct. 2227. In such cases, the federal Due Process Clause is implicated, and therefore federal habeas review is warranted.
. Because we reverse the convictions on this ground, we do not reach the other issues raised by petitioners, including sentencing claims.