In deciding whether to issue a preliminary injunction, courts today focus on three factors: the likelihood that plaintiff will ultimately prevail on the merits; the harm defendant will suffer if the injunction is wrongly issued; and the harm plaintiff will suffer if the injunction is wrongly denied. The idea is to account for the possibility that the court might err in its prediction on the merits. If wrongful denial would be particularly harmful and there is a real chance of wrongful denial, the court is more reluctant to deny. By contrast, if wrongful issuance poses the greater threat, the court is more reluctant to issue. This decision rule has intuitive appeal but overlooks a key point: In most cases, the court will be just as uncertain about its estimates of the harms as it is about its prediction as to the outcome of the case. Thus, the conventional approach begins to unravel. A court cannot minimize the implications of its possibly errant prediction on the merits by blindly relying on its possibly errant estimates of relative harm. The optimal decision rule must account for both types of uncertainty.