The quality of knowledge base is important for knowledge intensive systems(KIS), such as expert systems and intelligent decision support systems. But over a long time, knowledge acquisition is a bottleneck in designing KIS due to the bad coordination between knowledge engineers and experts, which is limited by human epistemology level nowadays. To solve the problem, this paper utilizes some machine learning theories (mainly rough set theory) to get a rough knowledge base and then choose genetic algorithms,visualization techno logy, know ledge verification and validation technology to optimize it. We, in this way, build a new relation between knowledge engineers and experts in which experts play a decisive role while knowledge engineers are accessory, i. e. it is expert oriented during the know ledge acquisition course