OPINION
KELLER, J.,
delivered the opinion of the Court,
in which McCORMICK, P.J. and MEYERS, PRICE, HOLLAND, WOMACK, JOHNSON, and KEASLER, JJ. joined.
The issue presented by this case is whether our abolition of the juvenile exception to the accomplice witness rule, announced by Blake v. State, 971 S.W.2d 451 (Tex.Crim.App.1998), applies retroactively to cases currently pending on direct review or not yet final. We hold that our holding in Blake is retroactive and affirm the judgment of the Court of Appeals.
Appellant and his wife Josefina divorced in 1989. They were named joint managing conservators of their two children. In 1990, Josefina married John Garmon, and together they had a son. On May 3, 1996, appellant’s children, ages eleven and ten at the time, set fire to the home of their mother and stepfather. Josefina, John, and their son suffered injuries but survived. In March of 1997, appellant was convicted on three counts of attempted capital murder. According to the State’s theory of the case, appellant had used physical punishment and psychological pressure to manipulate his children into setting the fire -with the purpose of killing his ex-wife, her husband, and their son.
At trial, appellant submitted a written accomplice witness instruction concerning the two children and requested that the instruction be included in the jury charge. The trial court failed to include the instruction. At the time, the trial court’s failure to include the instruction was consistent with our cases holding that children were not covered by the accomplice witness rule. Villarreal v. State, 708 S.W.2d 845, 848-849 (Tex.Crim.App.1986), overruled, Blake v. State, 971 S.W.2d 451 (Tex.Crim.App.1998); see also Blake, 971 S.W.2d at 455-458 (cases discussed therein). However, we subsequently abolished [677]*677the juvenile exception to the accomplice witness rule in Blake.1
Before the Court of Appeals appellant contended, among other things, that the trial court erred in failing to submit his requested accomplice witness instruction. The Court of Appeals reversed and instructed that the case be remanded for a new trial.2 Although observing that the trial court’s actions were justified by the state of the law at the time, the Court of Appeals held that, in light of the subsequent holding in Blake, the trial court’s failure to submit the instruction constituted error. The Court of Appeals found that the error had been properly preserved and that some harm had been shown under standards set forth in Almanza v. State, 686 S.W.2d 157 (Tex.Crim.App.1984) and Saunders v. State, 817 S.W.2d 688 (Tex.Crim.App.1991).
In its petition for discretionary review, the State contends that Blake should not be applied retroactively. Relying upon Geesa v. State, 820 S.W.2d 154 (Tex.Crim.App.1991), the State contends that justice would be best served by giving the Blake rule limited prospectivity. The State contends that retroactive application of the rule would be inequitable because (1) two defendants tried on the same date could obtain different results depending upon the speed of an appellate court issuing an opinion, (2) retroactive application of the new rule increases the State’s burden of proof after the case is presented and closed, and (3) retroactive application would require a jury to be charged on law that was not in effect at the time of trial. The State also asserts that “it makes more sense to allow defendants in the cases pending on direct review or not yet final to have their cases decided under the old rule than to require new trials in every case for failure to charge the jury with an instruction on accomplice witness testimony.” We granted this petition to address the scope and effect of doctrines relating to the retroactivity of new rules.
I.
Under the early common law, an appellate court never created law, it simply “discovered” the law. Linkletter v. Walker, 381 U.S. 618, 623, 85 S.Ct. 1731, 14 L.Ed.2d 601 (1965). When an appellate court overruled one of its earlier decisions, that earlier decision “was thought to be only a failure at true discovery and was consequently never the law.” Id. Conversely, the new decision, being the “accurate” discovery of the law, was not new law, but was an application of what had always in fact been the law. MThe “discovery” view of judicial decision-making came under attack as failing to reflect the realities of an appellate court’s task. Id. An alternative view held that judges fill in the blanks left by statutory law with judicial interpretation that more clearly defines the law. Id. at 624, 85 S.Ct. 1731. Under this view, an earlier decision is not erased without a trace by a later, overruling one; instead, the earlier decision stands “as an existing juridical fact until overruled.” Id. Hence, the overruled opinion retained vitality as to cases decided before the overruling occurred. Id.
Against this backdrop of conflicting views of judicial decision-making, the Supreme Court developed a pragmatic, case-by-case approach for new constitutional rules. Id. at 629, 85 S.Ct. 1731. This approach was perhaps most clearly spelled [678]*678out in Stovall v. Denno, 388 U.S. 293, 87 S.Ct. 1967, 18 L.Ed.2d 1199 (1967), which applied a three-factor test for determining whether a new rule should be given retroactive effect:
(a) the purpose to be served by the new standards,
(b) the extent of reliance by law enforcement authorities on the old standards, and
(e) the effect on the administration of justice of a retroactive application of the new standards.
Id. at 297, 87 S.Ct. 1967. Nevertheless, the Court recognized that the holding in its cases would always apply to the parties in the case in which the rule is announced, because “constitutional adjudications [do] not stand as mere dictum.” Id. at 301, 87 S.Ct. 1967 According to the Court, “sound policies of decision-making” required that it “resolve issues solely in concrete cases or controversies,” and applying a new rule in the case announcing the rule is necessary to give counsel incentive “to advance contentions requiring a change in the law.” Id. Although acknowledging that some “[ijnequity arguably results from according the benefit of a new rule to the parties of the case in which it is announced but not to other litigants similarly situated in the trial or appellate process who have raised the same issue,” the Court regarded “the fact that the parties involved are chance beneficiaries as an insignificant cost for the adherence to sound principles of decision-making.” Id. In essence, if the factors weighed in favor of prospective application, Stovall required that the new rule be given “limited prospectivity”: the new rule would be applied to the parties in the case in which the new rule was announced (a retroactive application), but the new rule would otherwise be prospective, applying only to cases in which the operative facts occurred after the adoption of the new rule.3
But the Supreme Court later repudiated the Stovall doctrine of retroactivity and its underlying rationale. Griffith v. Kentucky, 479 U.S. 314, 107 S.Ct. 708, 93 L.Ed.2d 649 (1987). The Court criticized as arbitrary the practice of “simply fishing one ease from the stream of appellant review, using it as a vehicle for pronouncing new constitutional standards, and then permitting a stream of similar cases subsequently to flow by unaffected by the new rule.” Id. at 323, 107 S.Ct. 708 (quoting Mackey v. United States, 401 U.S. 667, 679, 91 S.Ct. 1160, 28 L.Ed.2d 404 (1971)(Harlan, J. concurring)). The Court also criticized limited prospectivity as inequitable, violating “the principle of treating similarly situated defendants the same.” Griffith, 479 U.S. at 323, 107 S.Ct. 708 (citing Desist v. United States, 394 U.S. 244, 268-259, 89 S.Ct. 1030, 22 L.Ed.2d 248 (1969)(Harlan, J. dissenting) and United States v. Johnson, 457 U.S. 537, 556 n. 16, 102 S.Ct. 2579, 73 L.Ed.2d 202 (1982)). Consequently, the Supreme Court held that a newly announced constitutional rule for conducting criminal prosecutions must be applied retroactively to all cases, state or federal, pending on direct review or not yet final when the rule was announced. Griffith, 479 U.S. at 328, 107 S.Ct. 708.
In arriving at its holding in Griffith, the Supreme Court rejected the “clear break exception” to retroactivity that existed under prior caselaw. Under that exception, “a new constitutional rule was not applied retroactively, even to cases on direct review, if the new rule explicitly overruled a past precedent of this Court, or disapproved a practice this Court had arguably sanctioned in prior eases, or overturned a [679]*679longstanding practice that lower courts had uniformly approved.” Id. at 325, 107 S.Ct. 708. The Supreme Court explained that the “clear break exception” created the same problems that generally attended the use of limited prospectivity. Id. at 327,107 S.Ct. 708.
. The Supreme Court has applied a different holding to collateral review. With some narrow exceptions, new federal constitutional rules do not apply retroactively to cases that became final before the new rule was created. Teague v. Lane, 489 U.S. 288, 310, 109 S.Ct. 1060, 103 L.Ed.2d 334 (1989). The Court held that finality interests are of overriding importance. Id. at 309, 109 S.Ct. 1060. Once a case becomes final, the purpose of a new rule is no longer a relevant question; instead, the only relevant question is whether applying the new rule will further the purpose for which the writ of habeas, corpus is made available. Id. at 306, 109 S.Ct. 1060. Because habeas corpus is designed as a tool to deter the violation of established constitutional standards, a habeas court “need only apply the constitutional standards that prevailed at the time the original proceedings took place.” Id.
The Supreme Court’s retroactivity analysis for federal constitutional errors is binding upon the states when federal constitutional errors are involved. James B. Beam Distilling Co. v. Georgia, 501 U.S. 529, 535, 111 S.Ct. 2439, 115 L.Ed.2d 481 (1991). Conversely, Griffith and Teague do not bind the states on the retroactivity of new rules under state law. Geesa, 820 S.W.2d at 164; Mason v. Duckworth, 74 F.3d. 815, 818 (7th Cir.1996); People v. Murtishaw, 48 Cal.3d 1001, 258 Cal.Rptr. 821, 827, 773 P.2d 172 (1989); State v. Royer, 436 N.W.2d 637, 640 n. 2 (Iowa 1989); Commonwealth v. Waters, 400 Mass. 1006, 511 N.E.2d 356, 357 (1987); Sexton, 580 N.W.2d at 410; People v. Mitchell, 80 N.Y.2d 519, 591 N.Y.S.2d 990, 993, 606 N.E.2d 1381 (1992); Cooper v. State, 889 P.2d 293, 308 (Okla.Crim.App.1995); Farbotnik v. State, 850 P.2d 594, 602 (Wyo.1993). Nevertheless, a state may choose to adopt federal retroactivity standards as a matter of state law. The question we face, then, is what standards should we adopt for determining whether a new rule of state law should be given retroactive effect? In answering that question, we look to Geesa and to cases from other jurisdictions (both federal and state).
In Geesa, we abolished the reasonable-hypothesis construct for measuring the sufficiency of the evidence in a circumstantial evidence case, and we created a definition of reasonable doubt and imposed upon trial courts the duty to instruct juries upon that definition. 820 S.W.2d at 161 & 162. We discussed Linkletter, Stovall, Griffith, and Teague in deciding whether to make these holdings retroactive. Gee-sa, 820 S.W.2d at 163-165. While stating our general agreement with the Supreme Court’s rationale in Griffith, we held that Griffith’s rationale did not apply to the present case. Geesa, 820 S.W.2d at 165. We observed that the “Griffith rationale was keyed to the ‘actual inequity’ that results when only one of many similarly situated defendants receives the benefit of the new rule.” Geesa, 820 S.W.2d at 165. In Geesa, the State is the party that bene-fitted from the new rule — based upon the reasonable hypothesis construct, the Court of Appeals reversed the case and remanded with instructions to acquit. Id. at 155. We reversed that order for acquittal by jettisoning the reasonable hypothesis construct. Instead, the defendant was required to settle for a reversal and new trial based upon the absence of a reasonable doubt instruction. Id. at 163. Hence, the defendant in the case in which the new rules were announced did not benefit in any meaningful way from the new rules. And because Geesa formulated two new rules — one which benefitted defendants and one which worked to their detriment — applying both new rules in a limited prospective fashion could not be said to work any detriment to similarly [680]*680situated defendants. Id. at 165. The similarly situated defendants may not receive the benefit of the reasonable doubt instruction, but they would receive the benefit of the reasonable hypothesis construct, which was not available to the defendants in Geesa.
Although holding that Griffith’s rationale was not implicated, we also stated that retroactivity “must be decided on a case-by-case basis.” Geesa, 820 S.W.2d at 165. We conducted a Stovall-type balancing inquiry to determine whether the new rules should be applied retroactively or prospectively. Geesa, 820 S.W.2d at 165. We pointed out that the rules were procedural in nature and did not confer any greater constitutional protections than existed before, and we observed that imposing the rules retroactively would greatly burden the administration of justice by requiring new trials in every case for failing to give the definition of reasonable doubt. MBy expressing general agreement with the Griffith rationale, holding that the rationale was not implicated, and conducting a Stovall analysis, the Geesa opinion left an unresolved ambiguity in our law of retroactivity: whether Geesa was a mere anomaly to an otherwise uniform application of retroactivity under Griffith or whether Stovall’s case-by-case approach applied generally to new non-constitutional rules.
The federal courts have not agreed on whether Griffith applies to new federal rules of non-constitutional origin. The First Circuit has held that Griffith applies, finding “nothing in Griffith, either in terms or purport, distinguishing between constitutional and statutory interpretations.” United States v. Lopez-Pena, 912 F.2d 1542, 1545 (1st Cir.1989), vacated, 912 F.2d 1552 (1st Cir.1990), reinstated, 944 F.2d 51 (1st Cir.1991), cert. denied, 501 U.S. 1249, 111 S.Ct. 2886, 115 L.Ed.2d 1052 (1991)(magistrate who was not authorized to preside over jury selection according to new interpretation of the statute). By contrast, the Third Circuit held that “Griffith should be confined to constitutional rules of criminal procedure and thus does not require retroactive application of new procedural decisions not constitutionally grounded.” Diggs v. Owens, 833 F.2d 439, 442 (3rd Cir.1987). That court upheld an earlier case finding that a Supreme Court interpretation of the Interstate Agreement on Detainers Act was . not retroactive. Id. The earlier case found non-retroactivity after conducting a Stovall balancing test. United States v. Williams, 615 F.2d 585, 592-593 (3rd Cir.1980). The United States Court of Military Appeals avoided the possible application of Griffith by announcing that its new interpretation of a section in the military code would be fully prospective, so that the holding did not apply even to the parties in the case in which the new rule was announced. United States v. Carter, 25 M.J. 471, 477 (CMA 1988). “We conclude that — at least where a constitutional holding is not directly involved — Griffith does not prevent fully prospective overruling, which does not distinguish in result between similarly situated appellants.” Id.
Most of the states that have confronted the retroactivity issue in the context of non-constitutional rules have adopted the Stovall factor approach. Some of these cases involved judicially made rules. Waters, 511 N.E.2d at 357, rehearing, 399 Mass. 708, 506 N.E.2d 859, 862-863 (1987)(denying retroactivity to judge-made rule that coercion of a confession by a private individual requires suppression); Sexton, 580 N.W.2d at 409-416 (Mich.)(prophylactic rule that police must inform suspect when retained counsel is available for consultation); State v. Hudson, 331 N.C. 122, 415 S.E.2d 732, 751 (1992)(elimination of perjury as an aggravating factor); State v. Gordon, 913 P.2d 350, 353-355 (Utah 1996)(judicially-made rule prohibiting a part-time prosecutor from representing a criminal defendant as defense counsel); Farbotnik, 850 P.2d at 597-603(Wyo)(right to a complete record in a criminal case imposed pursuant to [681]*681high court’s supervisory power). Others involved the new interpretation of a statute. People v. Carrera, 49 Cal.3d 291, 261 Cal.Rptr. 348, 370, 777 P.2d 121 (1989)(statutory right to privacy of a pretrial detainee); People v. Erickson, 117 Ill.2d 271, 111 Ill.Dec. 924, 513 N.E.2d 367, 374-375 (1987)(statutory right to waive jury sentencing); Mitchell, 591 N.Y.S.2d at 993-995, 606 N.E.2d 1381 (statutory right of defendant to be present when trial court questions prospective jurors about bias); Cooper, 889 P.2d at 307-310 (Okla.)(statutory right to presumption of innocence that had been found to bar a jury instruction on flight when the defendant had given no explanation for his flight). A minority of states has adopted Griffith for all new rules, regardless of whether the new rule is of constitutional origin. Smith v. State, 598 So.2d 1063, 1066 (Fla.1992); Royer, 436 N.W.2d at 640 n. 2 (Iowa); State ex. rel. Schmelzer v. Murphy, 201 Wis.2d 246, 548 N.W.2d 45, 49 (1996). While holding that Griffith did not apply to new non-constitutional rules, the Arizona Supreme Court applied a “fairness” standard that appears to be similar to Griffith ⅛ underlying rationale. State v. Gardfrey, 161 Ariz. 31, 775 P.2d 1095, 1097 n. 1 (1989).
We agree with the more widely held view that a Stovall analysis should apply to new rules of non-constitutional origin. Not only is the Stovall approach supported by our prior opinion in Geesa, as well as the caselaw of most other states confronting this retroactivity issue, but we have commonly distinguished between constitutional and non-constitutional rules in other respects. We apply a more exacting harm analysis for constitutional violations than for non-constitutional errors. See Tex.RApp. P. 44.2(a) & (b). And, non-constitutional violations are generally not cognizable on habeas corpus. Ex parte Sanchez, 918 S.W.2d 526, 527 (Tex.Crim.App.1996). Moreover, while we are bound by the Supreme Court on matters of federal constitutional law, we have the power to construe state law differently and have often done so when the divergent interpretation seems to be better supported by reason or more in line with the intent of state lawmakers. See Cain v. State, 947 S.W.2d 262, 264 (Tex.Crim.App.1997)(ex-cept for federal constitutional errors labeled by the United States Supreme Court as structural, no error is immune from a harmless error analysis); Hulit v. State, 982 S.W.2d 431, 435-437 (Tex.Crim.App.1998)(no warrant requirement under Texas Constitution despite similarities in language to the Fourth Amendment). We reject Griffith’s approach as too rigid. Fairly. assessing whether a new rule should be given retroactive effect requires examining the various interests in favor of and against retroactivity, and Stovall’s three factor balancing test accurately encompasses the competing interests involved. Hence, we adopt the Stovall balancing test for determining whether new rules of non-constitutional origin should be given retroactive effect.
II.
The next question we must confront is how the Stovall test applies to a given case. To even implicate the retroac-tivity question, the articulated rule in question must in fact be a “new” rule. Court-made rules based upon supervisory or inherent judicial power may be new if no previous decision of the court has articulated the rule. The Geesa reasonable doubt instruction, for instance, was a new rule because we had never before articulated that requirement. However, when a court interprets a statute, the fact that the interpretation has never before been articulated is not enough for the interpretation to be considered a new rule. For example, a statute may never have been interpreted before. A first time interpretation, even if unanticipated by the parties in the case, cannot be considered a new rule because, presumably, the Legislature intended that interpretation when it enacted the statute. The parties are not entitled to be governed by an erroneous interpretation of a statute [682]*682when an authoritative interpretation did not yet exist. For an interpretation to qualify as a new rule, the interpretation must have been preceded by an earlier inconsistent interpretation. And the earlier interpretation must be one that can be viewed as authoritative by the court issuing the subsequent interpretation. In essence, the new interpretation must be a “clear break” with the past as that concept is explained in Griffith. For example, the clear break requirement is fulfilled when a court of last resort overrules a prior decision, and in doing so, replaces a prior interpretation of the statute with an inconsistent interpretation. To be a true case of overruling, the newer interpretation cannot be based upon a legislative change in the statute in question. Instead, the newer interpretation is articulated because the original interpretation has been found to be inaccurate, or at least insupportable in light of factors other than a legislative change in the statute.
Once an appellate court has determined that a rule is a new rule, the court must apply the Stovall factors. Most courts addressing the first factor — the purpose of the new rule — have found the key question to be whether the new rule impacts the truth-finding function of a trial. If the new rule does impact the truth-finding function, then the purpose factor will generally outweigh the reliance and administration of justice factors and require a holding of retroactivity. Conversely, if the new rule does not impact the truth-finding function, then the state’s reliance and the burdens upon the administration of justice are generally sufficient to deny retroactive effect to the new rule. Williams, 615 F.2d at 593 (“because violations of IADA bear little on the defendant’s actual guilt or innocence of the offense charged, retroactive application will not correct or affect any of the truth-finding process at trial”); Gardfrey, 775 P.2d at 1097 (rule concerning burden of proof on insanity defense held retroactive); Donaldson v. Superior Court of Los Angeles County, 35 Cal.3d 24, 196 Cal.Rptr. 704, 713 & 713 n. 11, 672 P.2d 110 (1983)(Stovall test usually denies retroactivity to search and seizure holdings but affords retroactivity to decisions which implicate question of guilt and innocence); Carrera, 261 Cal.Rptr. at 370, 777 P.2d at 143 (“Decisions have generally been made fully retroactive only where the right vindicated is one which is essential to the integrity of the fact-finding process. On the other hand, retroactivity is not customarily required when the interest to be vindicated is one which is merely collateral to a fair determination of guilt or innocence.”); Erickson, 111 Ill.Dec. 924, 513 N.E.2d at 367 (purpose of the new rule factor is particularly concerned with whether or not the new rule enhances the truth-seeking process; retroactivity is not favored when the new rule is not shown to bear directly upon the jury’s truth-seeking role); Sexton, 580 N.W.2d at 413 (rule concerning right to be informed about counsel’s availability is not relevant to the ascertainment of guilt and does not implicate the integrity of the fact-finding process; hence the rule should be prospective); Mitchell, 591 N.Y.S.2d at 995, 606 N.E.2d 1381 (retroactivity not mandated when new rule relates only indirectly to the fact-finding process); Farbotnik, 850 P.2d at 603 (rule requiring a complete record was not “sufficiently significant with respect to fact finding at trial to warrant retrospective application of that rule”).
Determining whether a rule impacts the truth-finding function is important in part because of the fairness concerns announced in Griffith. When a new rule is unrelated to the truth-finding function of a trial, retroactive application of the rule to the parties in the announcing case is a necessary price to pay for a system that encourages parties to challenge bad rules: “there is a need to reward the party ... responsible for convincing the Court to overrule unsound precedent.” Proctor v. State, 967 S.W.2d 840, 845 n. 5 (Tex.Crim.App.1998) (ellipsis inserted). In a sense, the party procuring the new rule gains a [683]*683windfall, required to encourage challenges to bad precedent, that we are not required to extend to similarly-situated individuals — at least when there is reliance by the State and applying the new rule retroactively would add burdens to the administration of justice. But, when “the new rule goes to the heart of the truth-finding function, raising doubts about the accuracy of guilty verdicts in past cases,” concerns of fairness and equity become more important. Waters, 511 N.E.2d at 357.
In line with the prevailing trend in other jurisdictions, we find that the key question in a Stovall inquiry will generally center around whether the new rule significantly impacts the truth-finding function of the trial: the new rule should generally be retroactive if it does, non-retroactive if it does not. This proposition generally holds true even though the State relied upon the old rule and even though retroactive administration of the new rule may pose some burdens on the administration of justice. Ordinarily, when a new rule replaces an old one, the State will have relied upon the old rule. And ordinarily, there will be some added burden upon the administration of justice that attends retroactively applying the new rule. But there are also some mitigating considerations that come into play with these factors. First, as discussed earlier in this opinion, non-constitutional rules are generally not cognizable on habeas corpus. And, even if they were cognizable, the finality interests articulated in Teague would appear to bar the retroactive application of new non-constitutional rules upon collateral review. See United States v. Martinez, 139 F.3d 412, 417 n. * (4th Cir.1998), cert. denied, 525 U.S. 1073, 119 S.Ct. 807, 142 L.Ed.2d 667 (1999)(“it would be anomalous to apply new rules of statutory interpretation retroactively, but not new constitutional rules, when collateral relief for statutory errors is more circumscribed than for errors of constitutional magnitude”). Hence, the retroactive application of a new rule would be limited. Moreover, the harshness of a retroactive rule can sometimes be mitigated by requiring a different remedy than would ordinarily be accorded the rule in its prospective application. A party prevailing under a retroactive application of a new rule may not necessarily “obtain the same relief that would have been awarded if the rule had been an old one.” James B. Beam Distilling Co., 501 U.S. at 535, 111 S.Ct. 2439. Moreover, the rules regarding preservation of error and the rules pertaining to harm analyses also mitigate against any hardship suffered by the State from the retroactive imposition of a new rule. Even if the rule is retroactively imposed, a defendant may yet be denied relief on the basis that the error was not properly preserved or was harmless.
Hence, if a new rule significantly impacts the jury’s truth-finding function, the Stovall factors will generally balance in favor of retroactive application: the purpose of the new rule factor will generally outweigh the kinds of reliance and administration of justice concerns that are ordinarily present whenever a new rule is applied retroactively. The result may differ, however, if in a given situation, the State is found to have unusually strong reliance interests, or the burdens upon the administration of justice are shown to be unusually high.
Conversely, if the new rule does not significantly impact the jury’s truth-finding function, the Stovall factors will generally balance in favor of prospective application: ordinarily, concerns regarding reliance and administration of justice will be enough to tip the scales of the balancing test against retroactivity. The result may be different, however, if the State’s reliance interests and burdens upon the administration of justice are insignificant or wholly absent.
III.
An application of the above principles to Geesa shows the correctness of our refusal [684]*684to afford retroactive effect to the holding in that case. The reasonable doubt definition imposed by Geesa clearly relates to the truth-seeking function since the definition attempts to clarify the reasonable doubt standard that is utilized by a jury in deciding the case. There is at least some question about whether the definition impacts the truth-finding function because we cannot be certain that a given jury’s understanding of reasonable doubt, unaided by definition, is substantially changed by the definition. Nevertheless, assuming that the definition does in fact impact the truth-finding function, other considerations unique to the Geesa ease prevent that assumption from dictating the result. First, the significance of the purpose factor is undercut by the absence of actual inequity in the case. As noted above, Geesa traded one beneficial rule for another, so that defendants governed by the new' rules were not necessarily in a better position than those governed by the old rules. Further, the administration of justice factor was exceptionally strong: every case that was tried to a jury would have error in it for failing to give the reasonable doubt definition if the Geesa rules were applied retroactively. Moreover, the ⅛ struetion was required even absent a request, and the Court may well have believed that the absence of the definition would result in reversible error without regard to harm. A subsequent decision from this Court came to that very conclusion. Reyes v. State, 938 S.W.2d 718 (Tex.Crim.App.1996).4
The present case, however, provides a much more compelling case for retroactivity. Unlike in Geesa, no tradeoff is present here; the new rule is clearly more beneficial to defendants than the old one. And abolishing the juvenile exception to the accomplice witness rule significantly impacts the truth-finding function of the trial. In a case in which a juvenile is an accomplice, the new rule entitles the defendant to have the jury instructed that it may not convict the defendant on the basis of the juvenile’s testimony unless there is other evidence tending to connect the defendant to the offense. Blake, 971 S.W.2d at 455.5 This is much different from a rule which changes the time limits for a trial to begin or one which revises the requirements for a legal search. The accomplice witness rule is a prism through which the jury evaluates the strength of the evidence to convict and whether the State has proven the defendant’s guilt. While juries may have a common sense notion that accomplice witness testimony is suspect, the accomplice witness rule goes beyond such abstract notions by placing a straitjacket on the jury’s ability to convict on certain evidence. The jury cannot convict unless there is non-accomplice evidence that the jury believes that tends to connect the defendant to the offense. This is so regardless of how believable the accomplice witness may have appeared on the witness stand and regardless of any characteristics of a particular witness (such as youth) .that might, in the jurors’ minds, distinguish that person from the typical “suspect” accomplice witness.
Further, the accomplice witness rule imposes a sufficiency review that would not otherwise be conducted by appellate courts. Accomplice witness testimony can be sufficient to support a conviction under the legal sufficiency standard dictated by Jackson v. Virginia, 443 U.S. 307, 99 S.Ct. 2781, 61 L.Ed.2d 560 (1979). McDuff v. State, 939 S.W.2d 607, 614 (Tex.Crim.App.), cert. denied, 522 U.S. 844, 118 S.Ct. 125, 139 L.Ed.2d 75 (1997)(accom-plice testimony need not be corroborated [685]*685to be legally sufficient to show corpus de-licti ); see also Cathey v. State, 992 S.W.2d 460, 462 (Tex.Crim.App.1999)(“The accomplice witness rule is a statutorily imposed sufficiency review and is not derived from federal or state constitutional principles that define the legal and factual sufficiency standards”). Under state law, however, if the State fails to produce any non-accomplice evidence connecting the defendant to the offense, then the defendant is entitled to an acquittal on appeal. See Texas Code of Criminal Procedure, Article 38.17; Munoz v. State, 853 S.W.2d 558, 559-560 (Tex.Crim.App.1993). Prospective application seems especially inappropriate when the new rule changes what constitutes a failure of proof in the State’s case.
There can be no doubt that the State has relied upon the juvenile exception to the accomplice witness rule. But this reliance does not appear to be more than the reliance one would ordinarily expect of the State when an old rule is replaced with a new one. And, as the present case shows, there will be some new trials resulting from the application of this rule. But, unlike in Geesa, a cascade of widespread reversals is not expected. This case will affect only a portion of the total cases tried to a jury — only those cases in which a juvenile may be an accomplice. And the new rule will not provide relief on collateral review. Further, while jury charge error can never be completely defaulted, the failure to object will result in a harm standard that is more onerous to the defendant. Almanza, 686 S.W.2d at 171. And even when error is properly preserved, a harm analysis still applies. Id.; Saunders, 817 S.W.2d at 690. While applying the rule retroactively may provide some additional burden upon the administration of justice by requiring a few cases to be retried, this burden is not of the magnitude required to justify prospective application for a new rule that impacts the truth-finding function.
We find that the Stovall balancing test supports applying Blake retroactively. Consequently, we hold that Blake applies retroactively to all cases pending on direct appeal or not yet final when Blake was decided. The judgment of the Court of Appeals is affirmed.
MANSFIELD, J., dissented with note.