Abstract: To improve the capability of acquiring knowledge in expert system,in this paper we have proposed a rough set based expert system model.This model introduces a knowledge filter during the phase of knowledge acquisition.According to the variation of the knowledge dependency,it evaluates and classifies the newly collected knowledge.This system also introduces a mechanism for knowledge reconstruction during building knowledge base to refine and restructure the primitive knowledge base building.This method presented in the paper not only removes the redundant attributes of the knowledge base,but also restructures the value space of the attributes,and improves the performance of the whole system significantly.