The Social Judgment Test was developed as a measure of a judge's ability to perceive individual differences in behavior and to apply these perceptions so as to recognize the individuals in other situations from their behavior. The behaviors used in the test were the written responses of a number of target individuals responding to the same situation, a complex simulation of administrative behavior. The test employed a matching procedure. Under these circumstances, in which the judge was forced to discriminate among targets, differences in the degree to which the judges were acquainted with the targets were eliminated, and all target individuals were playing the same role, in the same situations, mean scores were only somewhat (although highly significantly) above the chance level and the differences among Js in their ability to correctly identify targets were only minimally consistent. These findings were obtained in spite of the fact that an attempt was made to select maximally discriminable targets. Re-keying the test so that the most popular response was considered the correct one resulted in equally unreliable scores. Although it is possible that the results are a consequence of the specific testing format and materials used, we advance the hypothesis that individual differences in ability to predict the behavior of others, apart from situational factors, is of minimal importance to successful prediction.