A model for drug use with depressed patients was proposed to distinguish among active drug responders, placebo responders, and pill responders. A brief literature review was undertaken to provide a rationale for the choice of variables included in the model.Data from two multi-hospital collaborative studies were used to test the model's efficiency in predicting outcome to four classes of active drugs and a placebo: a major tranquilizer (chlorpromazine), a tricyclic antidepressant (imipramine), a minor tranquilizer (diazepam) and a hydrazine MAO inhibitor (phenelzine). The predictor variables included in the multiple regression analyses were sex, marital status, age, education, social desirability, number of prior psychiatric episodes, amount and type of prehospitalization medication and the psychiatrists' pretreatment ratings of severity of illness. The outcome measure in this study was the 3-week severity of illness rating by study psychiatrists.The results tended to support the proposed additive model of drug use. This model assumes that: a) there is a series of variables that predict response to either pill taking or hospitalization; b) there is an additional set of variables that predict response to an effective drug; and c) there is a third set of variables that predict response to a particular effective drug,i.e., are idiosyncratic predictors for that drug. To predict response to an ineffective drug or placebo, one would use the equations derived for pill response or hospital response. To predict response to an effective, active drug, however, this equation would have to be expanded to include all three sets of predictor variables.