The use of rough sets theory to select essential attributes that can represent the original data set is well known. Knowledge discovered from such essential attributes are typically represented as rules, and are therefore represen- tative of the original data. We present three results towards rule evaluation as an extension of the "Rules-as-Attributes measure". First, we present an approach of finding repre- sentative sets of rules for a given data set. Secondly, we suggest that the Johnson's Reducer of the ROSETTA soft- ware generates a reduct with the minimum number of rules, and can be considered as a minimum representation of the original knowledge. Our third result provides an integrated approach for rule evaluation based on both the Rule Impor- tance Measure and the method of finding representative sets of rules. We argue that this approach can take the represen- tative rules ranking into a further stage. These approaches are proposed to facilitate the rule evaluations and can pro- vide an automatic and complete comprehension of the orig- inal data set.