Methods by which rodent carcinogenicity can be predicted have been prospectively validated for 40 chemicals evaluated for carcinogenicity by the US National Toxicology Program. It is concluded that a chemical of unknown carcinogenicity can be predicted to be in one of three possible categories--probably carcinogenic, probably non-carcinogenic or of uncertain activity. The last category is unlikely to contain genotoxic trans-species and/or multiple-site carcinogens. The component parameters of such predictions are one or more of several aspects of chemical structure, genotoxicity and rodent toxicity. Each of these parameters requires refinement but all are developed to the point that they can be integrated to make assessment of possible carcinogenicity. Carcinogenicity tends to be overpredicted by this integrated technique, each part of which has already been simulated by computer modelling. Improvements in predictive methodology will flow from three assumptions: (i) that emphasis must be placed equally on the properties of the test chemical and the responses it elicits in tissues for which carcinogenicity is to be predicted, (ii) that the integration of different predictive technique is preferable to the exclusive use of a single technique, and (iii) that the general predictivity of any technique or combination of techniques appears to be limited to < or = 80%, imposed by inadequate knowledge, and uncertainties in the experimental evaluation and classification of carcinogenic responses for diverse chemicals. This last statement does not preclude the attainment of higher accuracy within a congeneric series of chemicals. Foreknowledge of the likely outcome of a rodent carcinogenicity bioassay is now possible and will contribute to the focusing of animal testing resources.