This paper describes inference architecture of an intelligent CAD system that is to be used for design tasks in which complex and large amount of knowledge is used. In order to investigate the issue of handling huge and complex knowledge, design knowledge is categorized into four types and inference methods for them are identified respectively: original design problems are decomposed into several modules of design processes and finally into sub-problems with the four types of knowledge as basic elements in a top-down manner. We demonstrate the implementation of the architecture with the result of the machining center - a sort of machine tools - design.