In this paper, a self-learning system based on objective used in fault diagnosis expert systems is presented It depends on the deep knowledge model of the diagnosed system and can improve diagnostic capability by expanding and satisfying the shallow knowledge base. Algorithm and principle of the self-learning system are described in details. As an application, the self-learning system has been embedded in a coal-cutter fauIt diagnosis expert system knowledge base and improve it. Generally, a learning process should satisfy conditions of (1) efficiency (the learning process should be quick enough to solve relevant problem), (2) accuracy (the learning process must have enough accurate solving means for a given problem) and (3) consistency (the learning result should be consistent with the original knowledge). In this paper, a self-learning system based on objective and its application in fault diagnosis expert system are presented.