I agree with Professor Rohlf. The issue for good clusters twofold: 1) using
an appropriate distance messaure (ie. Euclidean, Mahalanobis, "city-block") 
and 2) a clustering method: Wards, single-linkage, average linkage etc.

Many times it's try and error to find the right combination of the above. In 
for stock retruns, squared Euclidean distance with Wards produces nice result 
for cluster.

They key is to see and look at the clusters if they make sense. One could 
cluster data
use the cluster as seeds to a method as discriminant analysis and see how 
many miss-classifications occur.

Of course there are more robust clustering methods that detect outliers 
better and don't "lock-in" a case into a cluster. See the work by Rosseau and such 
cluster methods as Fanny.

Neil Gottlieb
ProfileDepot, Inc

CLASS-L list.
Instructions: http://www.classification-society.org/csna/lists.html#class-l