OPINION
CONNOR, Justice.These petitions for review raise various issues in a products liability case that has been before this court previously on appeal1 and on an earlier petition.2 The original trial in this case resulted in a jury verdict for the plaintiff. The case arose from an accident in which the driver of a Caterpillar-manufactured front-end loader was killed when the loader rolled over an embankment. In Caterpillar Tractor Co. v. Beck, 593 P.2d 871 (Alaska 1979) (hereafter “Beck I”), we reversed on several grounds and remanded for a new trial. At a pretrial conference numerous disputes arose, and it is from the lower court’s pretrial order that the parties petition for review. On January 9, 1981, we entered an order reversing in part and affirming in part the pretrial order, with this opinion to follow.
I
■ The first dispute concerns a proposed jury instruction involving the “scientific knowability” of a design defect. The instruction states:
“a) Some of the relevant factors to be considered in balancing the benefits of the challenged design against the risk of danger inherent in such design, are the mechanical feasibility of a safer alternative design, the gravity of the danger posed by the challenged design, the likelihood that such danger would occur, the financial cost of an improved design and the adverse consequences to the product and to the consumer that would result from an alternative design,
b) Where scientific knowability of the dangerous character of the product is an issue, it must be measured by the knowledge and information which existed when the product left the hands of the manufacturer. This is because where no indication of danger exists and no techniques for obtaining such information are available, a manufacturer has no basis for concluding that the product should not be marketed.”
Both Beck and Caterpillar argue that paragraph (b) should not be included in the jury instructions, although they advance differing reasons.
*792Beck argues that paragraph (b) should not be given because the “scientific knowability” of the danger to human life in the event of a Caterpillar Model 944 front-end loader rollover is not a genuine issue in this case. Caterpillar contends the instruction is erroneous because it is tantamount to imposing absolute liability on a manufacturer; a manufacturer should be liable, it argues, only if there is “objectively ascertainable knowledge and information available at the time of manufacture which would give notice of a defect.” It acknowledges, however, that the “scientific knowa-bility” instruction is incorrect “because it is an inappropriate test for the type of design defect at issue.”
As it is inappropriate to the facts of the case, we agree with the parties that the “scientific knowability” instruction should not be given.
The “scientific knowability” concept derives from our holding in Heritage v. Pioneer Brokerage & Sales, Inc., 604 P.2d 1059 (Alaska 1979). In that ease, a products liability action based upon strict tort liability, we held that it is a defense to strict liability when the risk of an injury was scientifically unknowable. In Heritage, the plaintiffs, buyers of certain mobile homes, argued that their new mobile home was dangerously defective because of the presence of formaldehyde fumes, which assert-edly caused certain personal injuries. On appeal, plaintiffs argued that the trial court erred in admitting evidence of the “scientific unknowability” of the risk involved. The evidence showed that exposure to the concentration of formaldehyde involved was “not known scientifically to cause permanent deep lung damage of the type suffered” by one of the plaintiffs. Id. at 1063. We approved an instruction allowing the jury to consider whether “the amount of scientifically knowable danger inherent in the product at the time it was sold to plaintiffs” outweighed the utility of the product, but we limited the appropriateness of the instruction to those cases “where the ‘know-ability’ of the dangerous character of the product is an issue.”3 Id. at 1064.
The instruction was appropriate in Heritage because there was an issue of whether it was scientifically knowable, at the time of manufacture and marketing, that the concentrations of formaldehyde involved could cause permanent deep lung damage. This case, however, presents dissimilar facts. Here, the decedent was killed when a front-end loader, manufactured by Caterpillar Tractor Co. rolled over an embankment and crushed him. Beck I, 593 P.2d at 874-75. Decedent’s wife and personal representative, Paula Beck, filed a wrongful death action alleging, in part, that the loader was defective because it was not manufactured with a roll-over protective shield (ROPS), which is an overhead protective canopy designed to withstand, and reduce the risk of a rollover, thus protecting the operator from being crushed. Id.
The particular risk of danger involved in this case was the risk that the front-end loader would roll over and crush the operator underneath. Unlike the Heritage facts, however, the “scientific knowability” of this risk is not an issue. At the time of design and production of the model of loader involved here, Caterpillar made a “deliberate decision” not to install any kind of protective canopy. 593 P.2d at 876. Such a decision could not have been made without some recognition of the danger of rollover. Moreover, that a person can be injured in a front-end loader that overturns is too apparent to reasonably raise a “scientific *793knowability” issue. Thus, the instruction should not be given.4
II
Caterpillar argues that the trial court also erred in ordering that evidence of post-manufacture accidents and design modifications will be admissible. The trial court held that such evidence is admissible toward: (a) the dangerous or defective condition of the 944 front-end loader; (b) causation; (c) the feasibility of alternative designs; and (d) to impeach various contentions of Caterpillar regarding the safety of the loader. Caterpillar contends that this ruling contradicts our statement in footnote 52 of Beck I, 593 P.2d at 886, to the effect that design defects are measured by the knowledge and information which existed when the product left the manufacturer’s hands. Thus, it argues, evidence of post-manufacture occurrences is irrelevant as a matter of substantive law and should be excluded under Alaska Rule of Evidence 402.5 or alternatively, such evidence is more prejudicial than probative and, therefore, should be excluded under Evidence Rule 403.6
Caterpillar places too much emphasis on the dictum contained in footnote 52. The footnote contains a somewhat unfortunate choice of words; its meaning should be limited to the principle that a substantial change in the product after it leaves the manufacturer’s hands will ordinarily defeat a claim based on strict tort liability. See W. Kimble & R. Lesher, Products Liability § 58 (1979).
The trial court’s ruling is consistent with Alaska Rule of Evidence 407,7 which provides that in a products liability action evidence of subsequent measures is admissible toward the feasibility of alternative designs as well as defective condition.8 We have consistently declined to rewrite the rules of practice from the bench, and decline to do so here.. See, e. g., Stores v. State, P.2d, Op. No. 2252 at 21-22 (Alaska, December 19, 1980); Buchanan v. State, 561 P.2d 1197, 1209 (Alaska 1977). Rule changes are more appropriately accomplished by amendment upon recommendation of the rules committee, the bench, and the bar.
Even so, Rule 407 is consistent with the majority of jurisdictions. It admits evidence that is highly probative of the exist-*794ence of a defect, which is the essence of a strict liability action for a defective product. The leading case allowing such evidence to be admitted, Ault v. International Harvester Co., 13 Cal.3d 113, 117 Cal.Rptr. 812, 528 P.2d 1148 (1975), is particularly persuasive in that Alaska’s law on design defects is taken from California.9 In Ault, the California Supreme Court was asked to extend to products liability cases the evi-dentiary rule precluding admission of post-injury accidents or design changes toward proof of negligence. After examining the underlying policies of this exclusionary rule, the court held it inapplicable in those products liability cases based on strict liability:
“[Cjourts and legislatures have frequently retained the exclusionary rule in negligence cases as a matter of ‘public policy,’ reasoning that the exclusion of such evidence may be necessary to avoid deterring individuals from making improvements or repairs after an accident has occurred.
When the context is transformed from a typical negligence setting to the modern products liability field, however, the ‘public policy’ assumptions justifying this evidentiary rule are no longer valid. The contemporary corporate mass producer of goods, the normal products liability defendant, manufactures tens of thousands of units of goods; it is manifestly unrealistic to suggest that such a producer will forego making improvements in its product, and risk innumerable additional lawsuits and the attendant adverse effect upon its public image, simply because evidence of adoption of such improvement may be admitted in an action founded on strict liability for recovery on an injury that preceded the improvement.”
Ault, 13 Cal.3d 113, 117 Cal.Rptr. 812, 528 P.2d at 1151-52. Moreover, in negligence actions the focus is on the defendant’s conduct at the time of the accident. Thus, it can be said that conduct after the occurrence is irrelevant. See Sappenfield v. Main Street & Agricultural Park Railroad Co., 91 Cal. 48, 27 P. 590, 593 (1891). In products liability actions based on strict liability, however, the policy considerations which are involved shift the emphasis away from the manufacturer’s conduct to the character of the product. Evidence about the character of the product, as reflected in subsequent modifications or accidents, is highly probative in strict liability cases, and therefore is not excludable on relevancy grounds.
Other jurisdictions have reached a similar conclusion. See Robbins v. Farmers Union Grain Terminal Association, 552 F.2d 788, 793 (8th Cir. 1977); Roberts v. May, 583 P.2d 305, 309 (Colo.App.1978); Rucker v. Norfolk & Western Railway Co., 64 Ill.App.3d 770, 21 Ill.Dec. 388, 381 N.E.2d 715, 725 (1978), rev’d on other grounds, 77 Ill.2d 434, 33 Ill.Dec. 145, 396 N.E.2d 534 (1979); Ginnis v. Mapes Hotel Corp., 86 Nev. 408, 470 P.2d 135, 140 (1970); Barry v. Manglass, 55 A.D.2d 1, 389 N.Y.S.2d 870, 874-75 (1976); Shaffer v. Honeywell, Inc., 249 N.W.2d 251, 257 n.7 (S.D.1976). Cf. Bolm v. Triumph Corp., 71 A.D.2d 429, 422 N.Y.S.2d 969, 973-74 (1979) (post-occurrence evidence inadmissible in design defect products liability case because substantive standard of liability resembles negligence standard). Contra, Price v. Buckingham Manufacturing Co., 110 N.J.Super. 462, 266 A.2d 140, 141 (1970); Haysom v. Coleman Lantern Co., 89 Wash.2d 474, 573 P.2d 785, 790-91 (1978); Comment, Death Knell to the Exclusionary Rule Against Subsequent Remedial Conduct in Strict Products Liability, 13 San Diego L.Rev. 208 (1975). See also Davis, Evidence of Post-Accident Failures, Modifications and Design Changes in Products Liability Litigation, 6 St. Mary’s L.J. 792, 796-801 (1975); Note, Products Liability and Evidence of Subsequent Repairs, 1972 Duke L.J. 837, 845-50; Note, 44 U.Cinn.L.Rev. 637 (1975).
We therefore hold that evidence of post-injury accidents and design changes is admissible as authorized by the trial court.
*795III
Finally, Caterpillar seeks review of the trial court’s order that Beck’s possible negligence be relitigated. In Beck I, 593 P.2d at 890, we held that Beck’s negligence, if any, in operating the loader would reduce his recovery, even though the action is based on strict tort liability. In the original trial the jury found Beck to be 50 percent negligent.
On remand, the trial court ordered that Beck’s negligence be relitigated, along with the issue of Caterpillar’s liability. Caterpillar does not appear to argue that the 50 percent finding should stand, but rather that the jury should be required to abide by the original jury’s determination that Beck was contributorily negligent to some degree, i. e., instruct the jury that Beck was negligent yet have them determine the extent.
What issues are to be relitigated upon remand, in the absence of a directive from this court, is within the discretion of the trial court. See State v. Kaatz, 572 P.2d 775, 784-85 (Alaska 1977) (sustaining trial court’s decision to allocate, itself, portions of liability based on evidence from the first trial while ordering relitigation of damages); Dowling Supply & Equipment, Inc. v. City of Anchorage, 490 P.2d 907, 910 (Alaska 1971) (sustaining trial court decision to relitigate just damages). See generally Alaska R.Civ.P. 59. Thus the inquiry is whether the trial court abused its discretion in ordering that both parties’ liability be fully relitigated. We conclude that it did not.
Partial new trials are most common where the issues can be separated. See Poulin v. Zartman, 542 P.2d 251, 275-76 (Alaska 1975). The most typically separable issues are those of liability and damages. See Sturm, Ruger & Co. v. Day, 615 P.2d 621, 623-24 (Alaska 1980); Dowling Supply & Equipment, Inc., 490 P.2d at 910. Where two issues are interwoven and not reasonably divisible, however, a new trial on both issues is required. See, e. g., Liodas v. Sahadi, 19 Cal.3d 278, 137 Cal.Rptr. 635, 562 P.2d 316, 320 (1977); Bassett v. O’Dell, 178 Colo. 425, 498 P.2d 1134, 1135 (1972); Shere v. Davis, 596 P.2d 499, 500 (Nev.1979). Whether the issues are sufficiently separable to warrant a partial new trial depends on the facts and circumstances of each case. Sturm, Ruger & Co., 615 P.2d at 623; see Poulin, 542 P.2d at 276.
We do not believe that the issue of defective design is sufficiently separable from that of Beck’s possible negligence to require a partial retrial. Beck’s possible negligence in the knowing use of a defective product is tied to the existence of a defect and a known risk.10 In a personal injury action, a new trial should ordinarily be had on all contested factual issues, “regardless of the ability of the parties on appeal to pinpoint error so as to show that the error ... may have affected only one issue.” Maxwell v. Portland Terminal Railroad Co., 253 Or. 573, 456 P.2d 484, 486 (1969). Further, Caterpillar’s suggestion that the second jury be told of and bound by the original jury’s determination of liability, but then be asked to determine the extent of liability, would be unduly prejudicial to Beck. Bassett, 178 Colo. 425, 498 P.2d at 1135; Juvland v. Mattson, 289 Minn. 365, 184 N.W.2d 423, 425 (1971). The general issue of liability frequently cannot be separated from the issue of percentage of liability attributable to each party. See Begin v. Weber, 234 N.W.2d 192, 195 (Minn.1975); Juvland, 289 Minn. 365, 184 N.W.2d at 425.
A partial retrial should not be ordered unless it clearly appears that the issue to be retried is distinct and separable, and can be had without injustice to any party. Sturm, Ruger & Co., 615 P.2d at 624; Juvland, 289 Minn. 365, 184 N.W.2d at 425. Neither of these conditions are met here. Thus we conclude that the trial court *796did not abuse its discretion in ordering a new trial on all issues of liability.
This case is remanded to the superior court for further proceedings consistent with this opinion.
MATTHEWS, J., not participating.. Caterpillar Tractor Co. v. Beck, 593 P.2d 871 (Alaska 1979).
. In Beck v. Caterpillar Tractor Co., Supreme Court No. 5165 (Alaska, March 14, 1980), we summarily reversed a ruling by the trial court denying Beck access to certain accident reports possessed by Caterpillar.
. We noted that a determination of the scientific knowability of the danger inherent in the product was relevant because it underlies an evaluation of the manufacturer’s ability to eliminate the danger. “This is because where no indication of danger exists and no techniques for obtaining such information are available, a manufacturer has no basis for concluding that the product should not be marketed.” Heritage, 604 P.2d at 1063 (footnote omitted). In other words, if the risk of the particular danger involved is scientifically unknowable when the product is marketed, the manufacturer generally cannot be held liable in strict liability.
. Even if appropriate, the instruction is inadequate in that it fails to tell the jury whether scientific knowability is an issue, it does not state the legal significance of the issue, and it does not indicate that the burden of establishing the defense of scientific unknowability is on the manufacturer.
. Alaska R.Evid. 402 provides:
“All relevant evidence is admissible, except as otherwise provided by the Constitution of the United States or of this state, by enactments of the Alaska Legislature, by these rules, or by other rules adopted by the Alaska Supreme Court. Evidence which is not relevant is not admissible.”
. Alaska R.Evid. 403 provides:
“Although relevant, evidence may be excluded if its probative value is outweighed by the danger of unfair prejudice, confusion of the issues, or misleading the jury, or by considerations of undue delay, waste of time, or needless presentation of cumulative evidence.”
Exclusion of prejudicial evidence under Rule 403 is discretionary with the trial court. No specific evidence has been proffered yet, and it ís therefore impossible to say there has been an abuse of discretion at this point.
. Alaska R.Evid. 407 provides:
“When, after an event, measures are taken which, if taken previously, would have made the event less likely to occur, evidence of the subsequent measures is not admissible to prove negligence or culpable conduct in connection with the event. This rule does not require the exclusion of evidence of subsequent measures when offered for another purpose, such as impeachment or, if controverted, proving ownership, control, feasibility of precautionary measures, or defective condition in a products liability action.”
. Even in negligence actions, where the rule of exclusion is most frequently applied, evidence of subsequent measures or accidents is generally admissible to impeach or prove that the product was capable of causing injury. McCormick on Damages §§ 200, 275 (2d ed. 1972). Thus such evidence is similarly admissible toward those issues in strict liability actions.
. In Beck I, 593 P.2d at 884-85, we adopted the defective design test enunciated in Barker v. Lull Engineering Co., 20 Cal.3d 413, 143 Cal.Rptr. 225, 573 P.2d 443 (1978).
. In Beck I, we held that:
“When the design defect is the lack of a safety device, the jury may be instructed that the plaintiff may be comparatively negligent in the knowing use of a defective product only if he voluntarily and unreasonably encounters the known risk.”
593 P.2d at 892.