It is shown that randomization applies in repeated measurements whenever the F-test applies in the analysis. Contrary to common practice, randomization otherwise invalidates the use of Hotelling's T2 and Box's approximate test, and it presents difficulties in testing that dispersion matrices have the structure .