This paper discusses the macroscopic and important problem in the field of KDD. First, it is very difficult to describe the complex type data by general knowledge representation method. So the authors use pattern, which is defined as the vector in Hilbert Space, to represent the characteristic of complex type data. It also can be used to describe the rule of knowledge discovery. Second, the general structure model is constructed based on complex type data—DFSSM (discovery feature sub-space model ) following by the research on inner mechanism of knowledge discovery system. At last, the authors prove the practicability and validity of this general structure model i. e. DFSSM which can guide the knowledge discovery of textual data and image data (meteorological echogram data). It will beapplied in other complex type data in future.