The cytokinesis block method was used to examine the intraclass correlation coefficient of the human lymphocyte micronucleus assay, sources of variability, and practical issues regarding the number of samples per subiect. Twenty samples of 100 binucleate cells from a single phlebotomy per subject were analyzed (n = 1 121, using methods to evaluate variance components. The results showed marked intraindividual (sampling error) variation greater than interindividual variation, and no be tweengroup contribution to the total variance. The intraclass correlation was 41.6%, indicating that slightly greater than half of the total variation in micronucleus outcomes was due to error variance (i.e., 58.4%). After adjusting for age, the intraclass correlation coefficient decreased trivially from 41.6% to 39.8%. There was a strong differential gender effect, favoring a greater micronuclei frequency in women. In conclusion, the data suggest that most of the variability in our data set for the micronucleus assay was due to sampling error; a strong differential gender effect favoring females was also verified. Equally important, in terms of practical applications, our analysis of the apprcpriate number of samples per subject revealed that scoring greater than 1,000 cells (1 0 determinations per subiect) yielded no substantial improvement in statistical sensitivity, compared to the traditional 20 determinations. We suggest that more attention should be directed toward improving the assay's utility, while reducing sampling error.