A Bayesian prediction strategy is outlined in which antecedent measures are divided into two subgroups. One subgroup is used to discriminate among criterion groups, the second to provide normal linear predictions for each group. Individualized regression constants are subsequently obtained by computing probabilities of group membership from the discriminating measures and weighting the group prediction equations by these probabilities. The technique is illustrated by the prediction of cumulative University of Washington GPA for student groups categorized by terminal university status using achievement and aptitude measures from the Washington Pre-College testing program. Errors on validation were slightly less for the adjusted predictions than for a single pooled prediction equation suggesting this may be a promising approach to the moderation of predictions. (Author)