The performance of mutation assays with single cells involves a series of separate steps beginning with the induction of mutant cells and ending with the counting of mutant and wild-type colonies. The variation among identically treated cultures is here modeled as arising from 3 sources: (1) the number of mutant cells surviving treatment, (2) the number of mutant cells sampled in steps of sampling and growth required in assays involving phenotypic lag, and (3) the number of mutant and nonmutant colonies actually observed. The arithmetical statements describing the expectation of variance from each step are presented and used to provide means to calculate an expected overall variance for typical experiments. The model is then tested by comparing its predictions with the observed mutant fractions in human lymphoblastoid cells at the loci coding for 6-thioguanine, ouabain, podophyllotoxin, and 5,6-dichlororibofuranosyl benzimidazole resistances. The model is found to have excellent predictive qualities and should be useful in experimental design of studies involving induction of rare variants in single-cell systems.