Our work in the drug treatment of personality disorders has led to several conclusions. Drug therapy has a definite, but limited place in the treatment of these conditions, and drug-diagnosis interactions are now clearly established for both global clinical outcome and discrete qualitative outcome categories. The consistently beneficial effects of chlorpromazine in emotionally unstable personalities and imipramine in “pseudoneurotic” patients are contrasted with the lack of drug responsivity in hysterical and passive-aggressive patients. Although the present investigation supports the predictive utility of diagnosis, the problem of inter-judge unreliability is serious and requires the development of new, objective diagnostic methods.An empirical, decision tree method of generating psychiatric diagnoses by computer is described. The method makes no assumptions about univariate or bivariate distribution characteristics and is consequently free of some of the criticisms levelled at multiple-discriminant function analyses of psychiatric rating scale data. The present data support the validity of these “machine diagnoses” and show drug-diagnosis interactions of a similar magnitude to our criterion, clinical diagnoses.