According to a multiplicative model for prediction of cancer risk for genotoxic agents the incremental cancer risk is, for low-intermediate exposures, proportional to target doses of the genotoxic substance and to the background risk in control groups. This model has been applied to evaluate cancer tests of acrylamide in rodents. Because of its reactivity toward DNA, glycidamide is assumed to be the causative genotoxic metabolite of acrylamide. Evaluation of experimental data according to the multiplicative model shows that mice, compared with rats, are of the order of 10 times more sensitive per administered dose of acrylamide. The US EPA procedure would, however, generally predict rats to be about twice as sensitive as mice to carcinogenic chemicals, because their estimates are based on scaling of the dose per square meter body surface area, as a surrogate for metabolic differences between the species. The comparison of rats and mice with respect to observed cancer incidence is at a key position in the evaluation of the usefulness of risk models for extrapolation between species. In the present study mice and rats were compared, with respect to in vivo doses of acrylamide and the metabolite glycidamide, after exposure to acrylamide. The relative in vivo doses were inferred from levels of hemoglobin adducts. The adduct levels from glycidamide were, per administered dose of acrylamide, approximately 3-10 times higher in mice than in rats. In combination with the above mentioned higher sensitivity of mice than rats in cancer tests of acrylamide this is compatible with the concept that glycidamide is the key genotoxic factor in acrylamide exposure. Furthermore, it is shown that the multiplicative, i.e. relative, risk model and measurements of the dose of the genotoxic factor give good prediction of the observed risk from acrylamide in cancer tests with rats and mice.