Opinion by Judge FERNANDEZ; Dissent by Judge REINHARDT.
FERNANDEZ, Circuit Judge:Joe Leonard Lambright and Robert Douglas Smith were found guilty of murder and sentenced to death. Their convictions and sentences were affirmed by the Arizona Supreme Court,1 and they, ultimately, sought federal writs of habeas corpus on various grounds. The district court denied the writs and they appealed. A three-judge panel of this court ordered that the writs be granted on the single ground that they were deprived of their constitutional rights when the Arizona trial court used dual juries (one for each of them). See Lambright v. Stewart, 167 F.3d 477 (9th Cir.1999). It did not decide the other issues that they raised. We withdrew the panel’s opinion and ordered that the issue be heard en banc. See Lambright v. Stewart, 177 F.3d 901 (9th Cir.1999). Because we disagree with the panel’s determination, we now affirm the district court on the dual jury trial issue.
BACKGROUND
Lambright and Smith were traveling across the country with Lambright’s girlfriend, Kathy Foreman. Smith was troubled by the fact that while Lambright and Foreman had intercourse in his presence, he did not have anybody along to satisfy him. For his part, Lambright thought that he “would like to kill somebody just to see if he could do it.” Lambright I, 138 Ariz. at 66, 673 P.2d at 4. They decided that both desires could be fulfilled, and they set out with Foreman to find a victim. They found Sandy Owen and kidnaped her. Smith raped her on the way to a mountain site where they all got out of the car and Smith raped Owen again as Lambright and Foreman had intercourse. What happened next was that Smith began choking Owen, and Lambright declared that she must be killed. So, “Lambright took Foreman’s knife out of its sheath and began stabbing the victim in the chest and *1183abdomen, twisting the knife around inside of her. Smith held one of the victim’s arms while she was being stabbed, and Foreman held the other arm.” Id. at 67, 673 P.2d 'at 5. After that, “Smith unsuccessfully tried to break Ms. Owen’s neck by twisting her head. Then Lambright, Foreman or both began cutting deeply into the victim’s neck with the knife.... The victim remained alive, and was at least semiconscious, as she attempted to raise herself up on one arm. Lambright picked up a large rock and hurled it at her head. Foreman testified that as he threw the rock he yelled ‘Die, bitch.’ ” Id. The three then drove off in a celebratory mood, playing the piece “We Are the Champions” as they went. See id. Once caught, the trio’s song changed. Foreman turned state’s evidence, was given immunity, and testified against her erstwhile lover and his friend. Lambright confessed, but deemed Smith to be the worst of the three. Smith, too, confessed, but he dubbed Foreman and Lambright as the real killers.
Lambright and Smith were joined in a single indictment because of their jointly facinorous conduct, but “[i]n light of the defendants’ confessions, which were not totally interlocking, and the appearance of potentially antagonistic defenses, [the trial judge] severed the cases of Lambright and Smith.” Id. As the Arizona Supreme Court pointed out:
Because most of the evidence was relevant to both defendants, however, the judge decided to hold a single “dual jury” trial, in which two separate juries were empaneled, each to decide the guilt or innocence of only one defendant, and each permitted to hear only evidence admissible against that one defendant.
Id.
The Arizona Supreme Court then held that the trial judge had erred when he resorted to the use of dual juries. See id. at 69, 673 P.2d at 7. However, it also determined that no prejudice had been shown. See id. at 70, 673 P.2d at 8. Lam-bright and Smith now ask us to overturn the district court’s decision and their convictions because, as they say, the use of the dual juries violated their due process rights.2
JURISDICTION AND STANDARD OF REVIEW
The district court had jurisdiction pursuant to 28 U.S.C. § 2254. We have jurisdiction pursuant to 28 U.S.C. § 1291. ‘We ... review a district court’s decision to grant or deny a § 2254 petition de novo.” Smith v. Stewart, 140 F.3d 1263, 1268 (9th Cir.1998).
DISCUSSION
Lambright and Smith suggest that the fact that the state trial court violated state procedural rules should lead to a decision that their due process rights were violated in this case, and that, indeed, dual jury use in their circumstances is so unreliable that the United States Constitution must have been violated. We disagree.
An error of state procedure is not, ipso facto, federal constitutional error. See Estelle v. McGuire, 502 U.S. 62, 112 S.Ct. 475, 116 L.Ed.2d 385 (1991). Rather, as the Supreme Court has told us with some asperity:
We have stated many times that “federal habeas corpus relief does not lie for errors of state law.” Today, we reemphasize that it is not the province of a federal habeas court to reexamine state-court determinations on state-law questions. In conducting habeas review, a federal court is limited to deciding whether a conviction violated the Consti*1184tution, laws, or treaties of the United States.
Id. at 67-68, 112 S.Ct. at 480 (citation omitted); see also Gerlaugh v. Stewart, 129 F.3d 1027, 1032 (9th Cir.1997); Christian v. Rhode, 41 F.3d 461, 469 (9th Cir.1994).
In this case there may or may not have been a state procedural error. In Lambright I, 138 Ariz. at 67-70, 673 P.2d at 5-8, the Arizona Supreme Court said that the trial court had erred when it used dual juries. That was so because the court felt that “local rules must first be approved by this court....” Id. at 69, 673 P.2d at 7 (citation omitted). Thus, because the court had not specifically authorized the use of dual juries, that use was deemed to be inappropriate. However, on further reflection, the Arizona Supreme Court has decided that the use of dual juries is not even an Arizona procedural error and has actually overruled Lambright I. As it said in Hedlund v. Sheldon, 173 Ariz. 143, 146, 840 P.2d 1008, 1011 (Ariz.1992), “‘[t]rial judges have inherent power and discretion to adopt special, individualized procedures designed to promote the ends of justice in each case that comes before them.’ ” Id. (citation omitted). Therefore, “[t]he trial judge’s decision to empanel a dual jury was not a local rule, and thus the judge’s implementing order did not exceed his authority. ...” Id. Just how that change of heart should be treated for purposes of this case need not be mooted. Even if we assume that Hedlund does not tell us what the law always was — that Lambright I was just a mistaken grasp at the true rule of law — the most that we have is a procedural error by the trial court. That does not demonstrate that, because jury use was involved, there was a federal constitutional error.
No doubt, under the Sixth Amendment to the United States Constitution a defendant in a criminal case has a right to a jury trial, but that does not even mean that a state is required to use the traditional twelve-person jury. Variations are permitted. See Williams v. Florida, 399 U.S. 78, 86 & 102-03, 90 S.Ct. 1893, 1898 & 1907, 26 L.Ed.2d 446 (1970). Nevertheless, there are limits. The state cannot reduce the number below six persons, see Ballew v. Georgia, 435 U.S. 223, 244-45, 98 S.Ct. 1029, 1041, 55 L.Ed.2d 234 (1978), and if the state goes that low, the determination must be unanimous. See Burch v. Louisiana, 441 U.S. 130, 138-39, 99 S.Ct. 1623, 1627-28, 60 L.Ed.2d 96 (1979); cf. Sochor v. Florida, 504 U.S. 527, 530, 112 S.Ct. 2114, 2118, 119 L.Ed.2d 326 (1992) (jury recommendation regarding death penalty not unanimous).
The lower limit is a recognition of the fact that “[a]t some point [a] decline [in numbers] leads to inaccurate fact-finding and incorrect application of the common sense of the community to the facts.” Ballew, 435 U.S. at 232, 98 S.Ct. at 1035. Accuracy of the result will be affected. See id. at 234, 98 S.Ct. at 1036. And verdicts vary as juries become smaller. See id. at 236, 98 S.Ct. at 1037. In short, the process will become so unreliable that the jury will not be functioning in a way required by the Sixth Amendment; that is to say, in essence there will not be a jury. So, while states can provide for different numbers of jurors (experiment if you will), they cannot tinker with jury size to the extent that what is left bears the appellation “jury” but is not even an allotrope of the traditional institution that we have come to know and revere.
That is far from saying that every experiment (jury or otherwise) leads to unreliability; indeed, many experiments lead to better and stronger institutions. In other words, “experiment” is not a pejorative word. As the Third Circuit said over forty years ago, “fair new procedures ... are allowable, although not traditional.” Byrne v. Matczak, 254 F.2d 525, 529 (3rd Cir.1958) (even though dispersal of jurors during deliberation was contrary to old common law, it was permissible); see also Chandler v. Florida, 449 U.S. 560, 578-79, 101 S.Ct. 802, 811-12, 66 L.Ed.2d 740 (1981); Galloway v. United States, 319 *1185U.S. 372, 388-91, 63 S.Ct. 1077, 1086-87, 87 L.Ed. 1458 (1943). Thus, while Lam-bright dubs the Arizona trial court’s use of dual juries an experiment, that does not advance the argument. It says nothing of the efficacy of that experiment or of its reliability.3 But that is the very point. If the dual jury trial did not make the result unreliable, he can hardly complain.
The use of dual juries may have been somewhat more unusual when Lambright’s trial took place than it is now, but that is far from saying that it was unreliable. Actually, the use was not all that great an innovation at the time. See Lambright I, 138 Ariz. at 67-70, 673 P.2d at 7-8. At any rate, we now know that dual juries are in wide use and that they have worked out just fine. In fact, we have accepted.the use of dual juries in noncapital cases. See Beam v. Paskett, 3 F.3d 1301, 1304 (9th Cir.1993); see also United States v. Sidman, 470 F.2d 1158, 1167-70 (9th Cir.1972). So have other federal courts. See Mack v. Peters, 80 F.3d 230, 235 (7th Cir.1996); United States v. Lebron-Gonzalez, 816 F.2d 823, 830-31 (1st Cir.1987); United States v. Lewis, 716 F.2d 16, 19-20 (D.C.Cir.1983); United States v. Hayes, 676 F.2d 1359, 1367 (11th Cir.1982); United States v. Rimar, 558 F.2d 1271, 1273 (6th Cir.1977); see also People v. Harris, 47 Cal.3d 1047, 1071-76, 767 P.2d 619, 633-37, 255 CaLRptr. 352, 366-370 (1989). We see no difference in using them in a capital case, and we resile from any suggestion to the contrary in Beam.4 The issue, again, is not one that should revolve around the fear of novelty; it is, simply, a question of “whether the procedure ... comports with the basic norm of due process.” Lewis, 716 F.2d at 20; see also Chandler, 449 U.S. at 579, 101 S.Ct. at 812.
In so stating, we do not denigrate the reflection that death is different from other penalties; it most assuredly is. See Ford v. Wainwright, 477 U.S. 399, 411, 106 S.Ct. 2595, 2602, 91 L.Ed.2d 335 (1986); Beck v. Alabama, 447 U.S. 625, 637-38, 100 S.Ct. 2382, 2389-90, 65 L.Ed.2d 392 (1980); Gardner v. Florida, 430 U.S. 349, 357-58, 97 S.Ct. 1197, 1204-05, 51 L.Ed.2d 393 (1977). Yet, we do not see how that helps Lambright’s case because it does not add to his argument that dual juries are unreliable, or that the ones used here were. In fact, his attacks on the particular use here point to possible problems, but do not point to any actual or inevitable harms, or to any particular inaccuracy fostered by dual jury use.
Interestingly enough, the harms pointed to by Lambright’s generalities are all harms that can conceivably occur in any joint trial, with or without dual juries. In joint trials without dual juries, defense counsel and defendants often wind up at the same counsel table. And jurors know that some evidence can come in against one defendant but not against another one; they often even hear the evidence, but are told to use it as to one defendant only. There might be some rub-off. And mistakes might occur. Usually, those factors alone do not even offer grounds for a severance.
We have expressed our assessment of those dangers by requiring that “[a] defendant seeking a reversal by reason of a district court’s denial of a motion to sever must establish that the prejudice he suffered from the joint trial was so ‘clear, manifest or undue’ that he was denied a fair trial.” United States v. Throckmorton, 87 F.3d 1069, 1071-72 (9th Cir.1996) (citation omitted). That is even true where the defenses are antagonistic. See id.; see also United States v. Mayfield, No. 98-50100, slip op. 9807, 9817 (9th Cir. *1186Aug. 26, 1999). It is also interesting that we have declared that “[i]n assessing whether joinder was prejudicial, of foremost importance is whether the evidence as it relates to the individual defendants is easily compartmentalized.... Central to this determination is the trial judge’s diligence in instructing the jury on the purpose of the various types of evidence.” United States v. Vasquez-Velasco, 15 F.3d 833, 846 (9th Cir.1994) (citations omitted). But dual juries help assure that very compartmentalization by keeping dangerous evidence away from the ears of the jurors for the defendant to whom it does not apply.5
Likewise, despite the dangers, the Supreme Court has lauded the benefits of joint trials for they “ ‘play a vital role in the criminal justice system.’ ” Zafiro v. United States, 506 U.S. 534, 537, 113 S.Ct. 933, 937, 122 L.Ed.2d 317 (1993) (citation omitted).1 They “promote efficiency and ‘serve the interests of justice by avoiding the scandal and inequity of inconsistent verdicts.’ ” Id. (citation omitted). Because of that, even where defenses are antagonistic, few cases have been reversed on that ground. See id. at 538, 113 S.Ct. at 937. Rather, an objecting defendant will not be entitled to a severance unless he shows that “there is a serious risk that a joint trial would compromise a specific trial right of one of the defendants, or prevent the jury from making a reliable judgment about guilt or innocence.” Id. at 539, 113 S.Ct. at 938. We are satisfied that the use of dual juries can actually palliate, rather than exacerbate, the risks of a joint trial. Particularly in a case like this one, where the main reason for a severance was one specific class of evidence, the use of dual juries can capture both the advantages of a joint trial and the protections of separate trials.
To put it another way, the problems alluded to in Beam are not endemic to dual jury trials. There we spoke of the additional complexity of a dual jury trial and of the additional likelihood of error. See Beam, 3 F.3d at 1304. Again, however, any joint trial can be as complex or, perhaps, even more complex, and error is just as possible.' It is true that problems can develop if the trial (or experiment) is run wrongly, but they are not inherent, or inevitable, in dual juries. For that reason, we upheld the use of dual juries in Sidman, 470 F.2d at 1170, although we said that rules for their use should be established. And even if Beam, 3 F.3d at 1304, were correct in its supposition that the likelihood of error will increase when dual juries are used, that is far from suggesting that an error has occurred. As the Supreme Court has pointed out, if there is a danger of prejudice in some cases, that still does not justify an absolute ban on a procedure. See Chandler, 449 U.S. at 574-75, 101 S.Ct. at 810. And if an error did occur, it could be corrected in the usual way. In most cases, the analysis of that question would be similar to the methods employed in reviewing severance cases. See Mack, 80 F.3d at 235. As it is, no special problems are shown to have developed in the cases at hand.
In fine, we cannot agree that any due process right of Lambright or Smith was violated when the Arizona trial court resorted to the use of dual juries, and neither of them has convincingly pointed to some other specific trial right which was compromised by that use.6
CONCLUSION
Because there is no per se constitutional error in the use of dual juries, either in general or in this case, and because no *1187other constitutional error developed from their use here, we reject the claims of Lambright and Smith that their convictions must be reversed on those grounds. We overrule any suggestion in Beam that there may be an inevitable constitutional barrier to the use of dual juries in capital cases, experimentally or otherwise. That being said, we need not decide whether a constitutional error would have been structural or would have been subjected to harmless error review. See Brecht v. Abrahamson, 507 U.S. 619, 629-30, 113 S.Ct. 1710, 1717, 123 L.Ed.2d 353 (1993).
In fíne, the district court did not err when it rejected those claims, and the illness found by the panel was, therefore, iatrogenic. We have heretofore withdrawn the panel’s opinion, and we now affirm the district court in this respect and return the case to the panel for its consideration of the other issues raised by Lambright and Smith.
AFFIRMED in part and returned to the panel.
. See State v. Smith, 138 Ariz. 79, 673 P.2d 17 (Ariz.1983); State v. Lambright, 138 Ariz. 63, 673 P.2d 1 (Ariz.1983) (Lambright I). Unless otherwise stated, or clear from context, all references to and holdings regarding Lambright shall include Smith.
. Because we reject the dual jury claim, we need not consider whether accepting it would create a new rule of constitutional law. See Teague v. Lane, 489 U.S. 288, 310, 109 S.Ct. 1060, 1075, 103 L.Ed.2d 334 (1989). Nor need we consider whether the state has waived application of the Teague doctrine. See Boardman v. Estelle, 957 F.2d 1523, 1534 (9th Cir.1992).
. Experiment is not defined by the word unreliability. It simply means that we do not necessarily know the outcome at first. But we may — often do — have a very good idea of what the outcome will be. College (even high school) students conduct "experiments” in science classrooms everyday, but the outcome is very predictable if they do the experiment correctly. Nor are even truly new scientific experiments unreliable per se. Rather, they may well turn out to be elegant successes.
. See Beam, 3 F.3d at 1303-04.
. The argument that each defendant’s jury will "necessarily speculate” about the evidence being heard by the other defendant's jury is itself rank speculation. See Harris, 47 Cal.3d at 1072, 767 P.2d at 634, 255 Cal.Rptr. at 367.
. As to the claim of procedural due process violations based on some kind of liberty interest created by Arizona law, nothing that they have pointed to shows, in any way, that the more traditional procedure for determining guilt — use of one juiy — conferred any sort of substantive right or guarantee that a single *1187jury deciding Lambright's case would not be working in tandem with Smith's single jury. Absent that demonstration of a substantive end protected by the Arizona procedure, it could not be a deprivation of a liberty interest of some sort when the Arizona Supreme Court determined that while the use of dual juries was error, no harm arose from the use of the procedure in question. In short, absent the creation of some substantive predicate governing decision making, a predicate which did not even arguably arise here, there could be no protected liberty interest. Therefore, no liberty interest was violated. See Kentucky Dep’t of Corrections v. Thompson, 490 U.S. 454, 460-63, 109 S.Ct. 1904, 1908-10, 104 L.Ed.2d 506 (1989); Dix v. County of Shasta, 963 F.2d 1296, 1299-1301 (9th Cir.1992); see also Chaney v. Stewart, 156 F.3d 921, 925-26 (9th Cir.1998); Bonin v. Calderon, 59 F.3d 815, 841-42 (9th Cir.1995); Moran v. Godinez, 57 F.3d 690, 698 (9th Cir.1994); Smith v. Sumner, 994 F.2d 1401, 1405 (9th Cir.1993).
. See Graham v. Johnson, 168 F.3d 762, 788 (5th Cir. 1999) ("[T]his court has rejected a claim such as that made by Graham that the execution of an innocent person, even where no constitutional violation has taken place, contravenes the Fifth, Eighth, and Fourteenth Amendments.”)