A question that frequently occurs when using hierarchical cluster analysis is - how many clusters is best?  ClustanGraphics now offers two procedures to test for the best number of clusters:
Best Cut
Bootstrap Validation

Best Cut examines the sequence of fusion values obtained in a hierarchical classification for a large step change, as an indication of a substantial loos of homogeneity.  The cluster partition immediately before a large step is then indicated as more valid than the one after the fusion.

Bootstrap Validation seeks to identify partitions that are furthest from random.  The sequence of actual fusion values obtained for the given data are compared with hierarchical clustering sequences obtained from a randomization of the same data or proximities.  When tested on randomly generated data, bootstrap validation usually finds no partitions that vary significantly from random, whereas with real data such as Mammals it finds valid partitions