Abstract
A procedure is proposed whereby R test statistics F=(F1F2…Fr)together with "randomly generated critical points" (C1C2…Cr) may be used to construct a simultaneous test for
a family containing R hypotheses. This procedure provides simultaneous tests having an exact prescribed type I error rate; the procedure does not require the distribution of F to be known. The simultaneous test is illustrated for making all pairwise comparisons in a one-way ANOVA model.