Until recently, the in vivo erythrocyte micronucleus assay has been scored using microscopy. Because the frequency of micronucleated cells is typically low, cell counts are subject to substantial binomial counting error. Counting error, along with inter-animal variability, limit the sensitivity of this assay. Recently, flow cytometric methods have been developed for scoring micronucleated erythrocytes and these methods enable many more cells to be evaluated than is possible with microscopic scoring. Using typical spontaneous micronucleus frequencies reported in mice, rats, and dogs we calculate the counting error associated with the frequency of micronucleated reticulocytes as a function of the number of reticulocytes scored. We compare this counting error with the inter-animal variability determined by flow cytometric scoring of sufficient numbers of cells to assure that the counting error is less than the inter-animal variability, and calculate the minimum increases in micronucleus frequency that can be detected as a function of the number of cells scored. The data show that current regulatory guidelines allow low power of the test when spontaneous frequencies are low (e.g., < or =0.1%). Tables and formulas are presented that provide the necessary numbers of cells that must be scored to meet the recommendation of the International Working Group on Genotoxicity Testing that sufficient cells be scored to reduce counting error to less than the inter-animal variability, thereby maintaining a more uniform power of detection of increased micronucleus frequencies across laboratories and species.