Polydiagnostic follow-up studies should explore whether certain definitions of a given disorder permit better prediction of the illness course than others and whether this good predictive validity is related to specific etiopathogenetic conditions. In order to carry out such studies successfully they should be based on broadly defined samples and comprise provisions for additional validation such as genetic data, neuropsychological testing etc. After an assessment at baseline and at discharge from hospital, the follow-up assessments should comprise five steps: (1) Identification of successful and unsuccessful diagnostic systems; (2) identification of features determining successful attribution; (3) analysis of successful systems; (4) analysis of unsuccessful systems, and (5) analysis of cases which have changed diagnostic attribution. The conclusions drawn from these analyses are intended to refine classification in psychiatry.