In Experiment I, forty-five schizophrenics and forty-five normals were randomly assigned to one of three conditions : information, non-social punishment and social punishment. A probability learning task with event ratio of 80 : 20 was employed. Punishment facilitated learning in schizophrenics but not in normals. Social and non-social punishment did not produce significant differences in response frequencies, but response latencies and questionnaire data were interpreted as evidence that social punishment produced more conflict and discomfort in schizophrenics than did the other two conditions. In Experiment 11, thirty-nine schizophrenics were randomly assigned to information, reward and punishment conditions. For the first IOO trials, the event ratio was 80:20, after which it was reversed to 20:80. After the shift, SS receiving punishment learned the new ratio significantly better than S s in the reward and information conditions. T h e results suggest that punishment is more effective than reward or information alone with schizophrenics.