Problems of making selection decisions among animals from several environmental groups with heterogeneous variance are discussed. Unless correction for heterogeneity of variance is made, animals will tend to be selected from the more variable groups, especially if selection is intense. This is an accurate and fair procedure only when the heterogeneity reflects real differences in heritability. It is shown that, if observations are expressed as deviations relative to the estimate of standard deviation from the group, the accuracy is high unless groups are very small.