As khown, the much used x² goodness-of-fit test has the disadvantage that the division into classes thereby extensively follows according to the user's arbitrariness. For an optimal choice of classes Mann and Wald [10] recommended the use of power function of the x² test. However, the restriction to the case of equiprobable classes and the high numbers of classes obtained thereby do not satisfy. By a more careful numerical analysis one can show that a much lower number of classes leads only to unimportant power losses. In this way, a thumb rule results for the practical man: At the level of significance = 0,05 choice 16 classes, independently of the alternative. Furthermore, the optimal class division is in general not the equiprobable one, it can be found by using of dynamic programming.