Hi, I've looked into ways of implementing this procedure, i.e. repeating the two-way ANOVA many times, scrambling the order of cases across the treatments, to produce a distribution of F ratios for each effect. This seemed a job for the 'boot' package. However, I'm not sure I'm doing an actual randomization test, as opposed to a bootstrap here. This is how I've coded the test (using the poisons data): ---<---------------cut here---------------start-------------->--- require(boot) "boot.lm" <- function(data, i) { mod <- lm(time ~ treat * poison, data=data[i, ]) anova(mod)["F value"][[1]][-4] # the F ratios for each effect } poisons.boot <- boot(poisons, boot.lm, R=1000, sim="permutation") ---<---------------cut here---------------end---------------->--- Is this the right way to ask for a randomization test using 'boot'? Thanks in advance. Cheers, -- Seb