The basic topology, knowledge representation and inductive learning of an intelligent expert system on rough sets are introduced in this paper. Based on the indiscernibility relation of system knowledge expression, a compatible algorithm is proposed to simplify the knowledge expressions and to induce the system decision rules. In order to extract the significant attributes from elementary knowledge and to reduce the dimensions of the system representation, inductive learning is also introduced by means of reasoning the dependency of knowledge between the teacher and the learner.