Even though specificity has been one of the most useful conflict resolution strategies for selecting productions, most existing rule-based systems use heuristic approximation such as the number of clauses to measure a rule’s specificity. This paper describes an approach for computing a principled specificity relation between rules whose conditions are constructed using predicates defined in a terminological knowledge base. Based on a formal definition about pattern subsumption relation, we first show that a subsumption test between two conjunctive patterns can be viewed as a search problem. Then we describe an implemented pattern classification algorithm that improves the efficiency of the search process by deducing implicit conditions logically implied by a pattern and by reducing the search space using subsumption relationships between predicates. Our approach enhances the maintainability of rule-based systems and the reusability of definitional knowledge.