Log-linear models are fitted to sister chromatid exchange (SCE) scores in order to test the significance of the differences in SCE scores observed between individuals or between experimental treatments. The analysis is performed at the level of chromosome groups. In each single test all measurements from all chromosome groups, both from the control and from the experimental sets, are utilized. By proceeding in this way full use is made of all the available information on the SCE scores at the level of chromosome groups and the shortcomings of the classical Student-t and chi-square tests are avoided.