The methodological issues and uncertainties arising from use of occupational cohorts in epidemiologic studies of quantitative risk assessment (QRA) for carcinogens were discussed. Inadequate exposure information was considered to be the weakest point of most epidemiologic studies, with only crude estimates or a range of exposures available for QRA. Better monitoring of worker exposure was seen as improving QRA from cohort data. Extrapolation of target tissue dose from external exposures was considered valid only if a linear relationship existed between tissue and external doses. The difficulties of confounding factors in cohort studies was contrasted with randomization of exposures possible in animal studies. Effect modifications were seen as introducing additional uncertainties into the QRA. The requirement of extremely large cohorts to achieve the required statistical power necessary to satisfy government risk limits was viewed as an additional limitation of cohort studies. Lifetime follow ups, determination of induction periods and lag time were discussed as other obstacles in cohort studies not present in animal studies. The occurrence of selection bias was determined to reduce the relevance of cohort studies to certain population groups. The empirical statistical models used to determine QRA from occupational mortality studies were classified as linear models, in which exposure effects are additive, and multiplicative models, in which exposure produces multiple background effects. The selection of an appropriate model was discussed. The inconsistencies arising from use of external reference groups, such as standardized mortality ratios, were countered with the possible bias of utilizing internal controls. The use of biologically based models was illustrated. The need to extrapolate hazard rates derived from the statistical models into lifetime risk was discussed. The authors conclude that information from both animal studies and epidemiologic cohort studies should be taken into account in QRA.