Knowledge usable by knowledge based clustering comes from two primary sources: 1) Cluster distribution in feature space; 2) Domain specific knowledge. Examples of the later for the MR domain include (a) white matter symmetry, (b) the fact that white matter is contiguous along both sides of ventricular area, (c) the ``X'' shape of the ventricular area in the "center slice", etc. The knowledge is encoded into rules in the Clips expert system tool. Generally, knowledge extraction from cluster distribution is done by clustering a set of patterns, projecting these clusters into feature space, then matching a cluster with the class it represents. Since most data sets are multidimensional, various projections may help separate different clusters and give a better understanding of the overall distribution.