In their 1984 article, Priest and Klein show that a simple divergent expectations model of the decision to litigate leads to a plaintiff success rate at trial that approaches 50 percent as the fraction of cases going to trial approaches zero. However, an extensive empirical literature has documented that plaintiffs win far fewer than half of their cases. As Priest and Klein observe, this conflict between the predictions of the model and the empirical literature may be attributable to violations in the data of the assumptions behind the simple model. Based on data from 3,529 cases, we find that "multimodal" case characteristics associated with violations of these assumptions cause plaintiff win rates to deviate from the 50-percent baseline in the manner that simple law-and-economics models would suggest. In other words, among cases that conform more closely to the assumptions underlying the simple divergent expectations model, the plaintiff win rate is closer to 50 percent. The 50 percent rule is actually a limiting implication of a selection effect that arises out of a simple divergent expectations model of the decision to litigate. In that model, each party estimates the quality of the plaintiff's claim with error, and the plaintiff settles when the defendant's offer is at least as large as the plaintiff's estimate of the value of her claim. Priest and Klein observe that cases selected for litigation are likely to be the difficult and uncertain ones-that is, the cases in which the true quality of the claim is close to the quality level needed for the plaintiff to win if the claim were to be tried-because the clear-cut cases will be more likely to settle before trial (or may never evolve into filed cases at all). The difficult and uncertain cases, in turn, are likely to be those that, on average, result in about half the victories going to one party and about half to the other.5 FEW results in the lawAlthough recent empirical work has validated the selection hypothesis, an extensive literature has documented that plaintiffs generally win far fewer than 50 percent of cases at the trial court or appellate level. This article investigates whether a "multimodal" approach to the selection of cases for litigation can reconcile the validity of the selection hypothesis in general with observed plaintiff win rates of less than 50 percent. The The theoretical motivation behind the 50 percent rule is as follows. If (C -S)/J is relatively high, say, 0.3,7 then cases that fail to settle will be those with P, > Pd, which will be disproportionately cases with Y' D-cases with the true quality close to the decision standard-because it is in those close cases that normal differences in estimation of quality across parties lead to large differences in the estimated probability of winning at trial. Put another way, if both parties agree that the plaintiff has a very small chance of winning, for example, Y' << D, then differences in parties' assessment of case quality are unlikely to be large enough to prevent the parties fr...