Psychiatry has been, among medical disciplines, the one showing the highest conflict between a «positivist» model and the «hermeneutic» tradition. In particular, in recent years several authors have suggested that both the scientific method and a scientific mentality should be introduced in psychiatric practice and research. They claim that criteria like a strong correlation among clearly defined signs and symptoms, a high predictive value towards the outcome and therapeutical response, and a clearly defined relationship between the symptoms and etiology are necessary to accept a diagnositc entity. Since in the case of e.g. schizofrenia such requirements are not met, some authors even suggest that the concept of schizofrenia should be abandoned. The present paper aims to show that such problems are not unique to mental disease, but are common to several branches of medicine. For example, the definition and classification of cancer follow a «polythetic» criterion which is similar to the concept of «family resemblances». Also, diagnostic reproducibility is often low in other medical disciplines. The paper examines how such methodological problems could be dealt with using the tools of epidemiology.