Data on rodents exposed to carcinogens indicate that their tumor probabilities are proportional to effective concentrations of parent compound or metabolites at the target tissues. This proportionality suggests that observed nonlinear dose-response curves reflect dose-dependent kinetics between applied dose rate and effective concentrations. Therefore low dose extrapolation procedures that include pharmacokinetic data could improve extrapolation accuracy. To test such procedures, we simulated bioassay and pharmacokinetic "data." Then, ignoring the mechanisms generating the data, we used four extrapolation procedures to estimate tumor probability at a low applied dose rate. Two of the procedures use a pharmacokinetic model and simulated pharmacokinetic data, and two do not. The pharmacokinetic model used for extrapolation was only an approximation to the one used to generate the pharmacokinetic data. The procedures that include pharmacokinetics often performed better and never did much worse than those that ignore them, regardless of the relations used to generate the data, the amount of experimental error in the pharmacokinetic data, and the appropriateness of the pharmacokinetic and extrapolation models used. Moreover they performed substantially better when effective concentration and tumor probability were concave-up functions of applied dose rate.