Starting from falsely equal and falsely unequal rates of successive psychotherapies, the prediction configural frequency analysis is introduced for evaluating psychotherapeutic effects by stratifying patients. As stratificator variables, intelligence quotients and sex of patients are used paradigmatically to predict strata-specific therapeutic effects and to compare different therapies conclusively even if patients cannot be assigned randomly to therapies. Prediction types of success are defined by those stratification patterns which favour therapeutic success.