Error message with repeated measures two-way ANOVA: There was a divide by zero error.

 The whole point of repeated measures ANOVA is to take into account differences between the subjects or blocks. If the mean values for each subject are identical, the sum-of-squares for intrasubject variability equals 0.0, and the repeated measures calculations are impossible. Prism (up to 5.03 and 5.0c) does not present a useful message, but rather simply says, "There was a divide by zero error."

In this situation, the subjects all have identical means, and so there is no advantage to using repeated measures. Use ordinary ANOVA instead.

One situation where this comes up is if the data entered are not measurements but rather ranks. If the repeated measurements for each subjects are ranked, then each subject will have the same average rank. You've accounted for repeated measures by ranking, so there is no point in using repeated measures ANOVA.