A dose-response analysis was carried out with 2 independent data sets available for ethylnitrosourea-induced specific-locus mutations in spermatogonia of the mouse. It was assumed that the occurence of mutation is binomially distributed and maximum-likelihood procedures were employed to determine the appropriateness of 4 alternative models, Linear, Linear-Quadratic, Power, and Threshold, in describing the dependence of the binomial parameter on dose. For both data sets, the Threshold model yielded a far superior fit and the threshold dose was estimated to be between 34 and 39 mg/kg. These results are supported by the relatively inefficient response of ethylnitrosourea at lower doses in inducing DNA adducts. Relevant specific-locus mutation results in the mouse for low-dose fractionated treatment as well as the recovery of mutation mosaics indicate the threshold model to be an oversimplification. Rather than a threshold dose below which 100% of the induced DNA adducts are repaired, we propose that some DNA adducts which may eventually be fixed as a mutation persist through a number of repair-competent cell divisions and do not interfere with normal cell function nor do they induce a repair response before being eventually fixed as a mutation. We interpret the threshold response for ethylnitrosourea-induced specific-locus mutations to be due to a saturable repair process which at lower doses results in ethylnitrosourea being less efficient in inducing mutation. Once this repair process is saturated, a clear dose-related increase in the mutation rate is observed.