When using a data mining systems, there are many different aspects which decide its overall quality. As for all computer programs we want it to be fast and use a minimum of resources. Still, what is most important is that it finds the patterns existing in our database, and that it creates rules which reflect these. Our demands for a good system are sometimes conflicting: We would like a system to be able to classify correctly as many as possible of the objects in our domain. In addition we would like the rules to be as simple as possible, as long as they are still correct for our training set. Lounis and Bisson define five main categories in [LB91] for evaluation. We take these as a starting point for our discussion below, but use our own arguments and draw our own conclusions.