A number of procedures are investigated for testing differences in variability in seven population distributions. Some procedures, previously considered robust, are found to be robust only with certain sample sizes. Testing for kurtosis is found to be useful for increasing power providedN>60. For equal sample sizes, maximum power is found to result from the application of one of Levene's procedures with platykurtic distributions and the Brown‐Forsythe procedure with leptokurtic distributions. No procedure is found to be adequately robust for all populations with unequal sample sizes. However, several methods are determined to be adequate if run at half the nominal alpha leve