This paper proposes a design concept and an organization structure of expert system for machine design. The technique of expert system is integrated with that of CAD and machine learning. According to the features of design problem, the design state space is divided into layers, so that a complex problem can be resolved into a series of simple problems. Three methods of knowledge representation, namely, rule, frame, and rule/frame, are used in system. Rules and frames are controlled by files, so that the system can efficiently undertake inference and justification with knowledge base. An effective control strategy is suggested for solving the problem. The inference can do the commonly used pattern recognition and matching. A rough set is used as a machine learning technique in the task of knowledge acquisition for expert systems. The CAD part of the system can do selection of bearing dimensions and strength calculation of shaft. Besides, the system has friendly man-machine interactive environment, explanation function, learning function, knowledge base maintenance system and graphic base. The technique of expert system is integrated with that of CAD and machine learning. According to the features of design problem, the design state space is divided into layers, so that a complex problem can be resolved into a series of simple problems. three methods of knowledge representation, namely, rule, frame, and rule / frame, are used in system. Rules and frames are controlled by files, so that the system an effective control strategy is suggested for solving the problem. A inference can be the commonly used pattern recognition and matching. A rough set is used as a machine learning technique in the task of knowledge acquisition for expert systems . The CAD part of the system can do selection of bearing dimensions and strength calculation of shaft. Besides, the system has friendly man-machine interactive environment, explanation function, learning function, knowledge base maintenance system and graphic base.