In order to achieve TV program group recommendation, an approach based on fuzzy clustering is proposed for program classification in this paper. This paper firstly describes the XML based program description metadata representation, in which both textual and symbolic information is included. Secondly it presents the program feature extraction and presentation method. A program is defined as two vectors, one is based on term statistics implying what the program is about, and the other reflects broadcasting characteristics of the program. Then the classifying approach based on fuzzy clustering is proposed. The approach goes: normalizing original data, building fuzzy similarity matrix, and then clustering. The final fuzzy similarity matrix is constructed by combining two fuzzy similarity matrices calculated from two different aspects.