A key-frame is a representation of one shot's content in the video, which usually reflects the main elements of a scene. Therefore, key-frame extraction and video analysis technology is the basis of content-based video retrieval. In this paper, a key-frame extraction method based on combination of mutual information and fuzzy clustering is proposed. In this method, the key-frames can maintain time-series and dynamic information of the video. And also, the key-frames can be a good reflection of the main contents of the video. Finally, one key-frame extraction system for MPEG-4 video is designed, and experiments show that the key-frame extraction system can be a better representative of video content, and conductive to the realization of video analysis and retrieval.