Severe criticisms are made of the present state of computer-assisted c
lassification methods. The literature gives little clarity on which of
many available computer methods is best, and serious problems have co
me to light in cladistics. Classifiers used to expect a high standard
of help from computers but actual results now engender a growing crisi
s of confidence. Since well-made groups are crucial in all fields of s
tudy, ways of improving the resolving power of grouping methods are di
scussed. Evolutionary groups are often riddled with inconsistencies, m
aking them polythetic. The Uniter computer program, just released, pro
vides for this by clustering using homogeneity, which allows dilation
due to contexts of external allies. Uniter puts the clusters it reveal
s into sequences, and evaluates the quality of each cluster, giving a
guide to placing the boundaries of formal taxa. The program's interact
ive treatments for a variety of situations are outlined, including a n
ew method for de-weighting coarsely coded properties with few states.
It is concluded that these treatments are essential for high resolving
power and programs that lack them should not be used for classifying
evolutionary groups.