An experimental approach was taken to the question of dose-response curves for chemical carcinogenesis, using DNA damage as a biomarker. Female rats were give 13 different doses of 1,2-dimethylhydrazine (from 1.4 to 135,000 micrograms/kg) and the subsequent hepatic DNA damage was determined by the alkaline elution technique. DMH doses below 450 micrograms/kg did not significantly damage DNA; all DMH doses of 1000 micrograms/kg or higher damaged rat hepatic DNA (P < 0.05). In this study the x values (dose) ranged over five orders of magnitude and the y values (DNA damage) ranged 30-fold. Ten different regression models (linear, quadratic, cubic, power, and six nonlinear transition models) were compared in their ability to fit the experimental data. With respect to log transformed dose, the six nonlinear transition equations fit the data considerably better than the four power type of equations. A sigmoid model fit to the log transformed dose of 1,2-dimethylhydrazine had an r2 of 0.9979, a degree of freedom adjusted r2 of 0.9969, a F-statistic of 1,457, and a fit standard error of 0.50. With respect to untransformed dose, only three equations (sigmoid, cascade and gaussian cumulative) could creditably fit the DMH data. The experimental results are interpreted with respect to hormesis, use of log transformed dose, sigmoid dose-response models, thresholds of biological response and cancer risk assessment.