Partitioning Around Mediods then rpart to follow - is this sensible

The problem:  There are no a priori groupings to run a classification on

My solution:

This is a non-R code question, so I appreciate any thoughts.  I have 
used pam in the cluster package proceeded by sillohouette to find the 
optimum number of clusters on scaled and centered data.  I have followed 
this by a classification tree analysis with rpart to discern which 
variables drive the clustering on the original data.  Is this a sensible 
approach?
many thanks,

Stephen Sefick