In this paper we present a rule reasoning method, which is based on rough sets theory, for inducing rules from examples. The key idea of the method is that it combines a criterion of the dependency degree of attributes with decision makers’ priori knowledge to select attributes of objects. Especially, it uses a compound weights algorithm to perform a proper reduction owing to several reductions that each rule can have and select the most effective attribute subset. As a result, a practical and effective reduced knowledge rule set can be acquired.