• Problems, DSM-III diagnoses, and treatment interventions were coded from the problem-oriented medical records of an acute psychiatric hospital. Problems and diagnoses were entered into separate stepwise multiple regression analyses to test their ability to predict 24 treatment interventions. Significant predictors were then included in a combined analysis. Both DSM-III diagnoses and problems were robust and significant indicators of medication treatments. Both were better predictors of medication than were problems and DSM-II diagnoses in an earlier study. With regard to psychosocial interventions, problems were better indicators than were DSM-III diagnoses, although neither was robust. Both were less useful as predictors of psychosocial interventions than in the earlier study. Overall, the combination of problems and diagnoses was generally superior in predictive power to either alone. These results suggest that problem nomenclature should include DSM-III diagnoses. Alternatively, DSM-III should be revised to accommodate some of the greater specificity and range of a problem nomenclature.