Numerous studies have concluded that the judged probability of conviction for a crime is higher when information concerning a prior conviction is disclosed. This is even more true when the prior and present crimes are the same. This paper examines the argument that the influence of a hypothetical previous conviction on the judgment of predilection toward a hypothetical present crime is a continuous function of the degree of similarity between past and present crimes. The subjects were presented with all possible pairs of 10 crimes, the first being considered as a prior conviction and the second as a presently charged crime. The subjects judged the subjective likelihood that a person convicted of the first crime (C1) would, in fact, he of a mind and inclination to commit the second one (C2). Independent judgments of intercrime, undefined similarity also were obtained. The main experimental findings, as expected, were that the judged probability of C2. given C1, was: (a) greatest when C2 was a repeat of C1; (b) uniquely different for each C1; (c) predicted quite well by degree of intercrime similarity; (d) poorly predicted by crime seriousness values. Thus, the subject appears to make judgments of predilection on the basis of simple representativeness heuristics, which specify that certain crimes will be considered more likely if they are more semantically related to earlier ones. Such a mental device unfortunately would he systematically biased in courtroom settings because judgments of intercrime similarity are not influenced by real-world probabilities.