Experimentalists have developed 2 stage carcinogenesis models within which initiation and promotion can be phenomenologically defined. For humans, no comparable sequential exposure situations have been studied, and the concepts of initiation and promotion cannot be defined in the same way that they have been for animal experiments. Nevertheless, there are epidemiological studies of human cancer which provide information on the effect of age at exposure, cessation of exposure and level of exposure on cancer risk which can be used to classify carcinogens according to their apparent mode of action. For the recognized human liver carcinogens, the available epidemiological data are not yet sufficiently detailed to permit clear conclusions about their role in multistage carcinogenesis.