Since the microbiological revolution, most infectious diseases have been defined and classified according to an etiologic criterion, i.e. the identification of single, external necessary causes (for example, Mycobacterium for tuberculosis). This is not the case with cancer. Not only external necessary causes of cancer have not been identified, but also the morphological classification cannot be based on univocal criteria. Although neoplasia and anaplasia appear to be universal attributes of cancer, these events are only quantitative. Neoplastic growth can be fast or slow (development may take weeks or years), and tissue pathologies are difficult to detect from normal tissue in some cancers but are obvious in others. Common special properties of anaplasia appear to be concealed in the wide range of morphologies. In the absence of a coherent morphological definition, and of external necessary causes (such as bacteria for infectious diseases), a mechanistic definition could be adopted. However, unless molecular biology discovers specific mechanistic steps in carcinogenesis, which indicate the existence of necessary events in carcinogenesis, we cannot adopt a univocal (monothetic) definition of cancer. The alternative is to use a polythetic definition, according to Wittgenstein's model of a long rope twisted together out of many shorter fibres.