Has anyone used hclus to determine p-values of a subset of features that give good class separation after clustering ? What I mean is sampling a set of features and checking for class separation by clustering and then determining if the features that give the actual class separation are not random. Cheers ../Murli