P3 amplitudes were often found to be reduced in schizophrenics, but in varying degree. We studied in a guessing task whether variations of group differences could even be obtained within the same task, by measuring P3 in the potentials evoked by the three qualitatively different events that occurred in each trial. We hypothesized that such variations were due to variations of subjective task relevance associated with each event. In each trial, a light or a sound was presented. Subjects had to confirm this 'primary event' by a delayed response, and then the current amount of money earned by correct responses was displayed. In the certain condition, the primary event was preceded by the appropriate cue on the screen ('light' or 'sound') whereas in the uncertain condition, the word 'guess' appeared. The task-relevance hypothesis predicted that schizophrenics would have smaller P3s than the control group with the primary event in the uncertain condition, but that the groups would not differ for the P3s evoked by the other events (certain primary event, cue and earnings display in uncertain and certain conditions). Essentially, this predicted pattern of results was obtained, but additional assumptions are needed for the task-relevance hypothesis to account for the precise pattern of results. Analysis of subjects' guessing behavior showed that control subjects' guesses were affected by the outcome of their preceding guesses while schizophrenics' guesses were not. This result provides some additional support for the hypothesis that the guessing task is less relevant to schizophrenics than to control subjects.