Ninety-eight psychology graduate students were asked to forecast first-yeargraduate grade point averages based on 110 profiles containing 10 cues derived from their fellow students' records. For each subject, the cue values were regressed onto the subjects' judgments (the model of man). Results indicated that the models (predicted judgments) were more accurate than the subjects' own judgments when correlated with the actual criterion. The average judgments for each profile were considerably more valid than the average of the judges' individual validities. Even the average judgments were improved upon by modeling. It was recommended that the model of man substitute for man in forecasting graduate grade point averages.