Abstract: MPEG-4 facial animation is one of the key technologies in the next generation model based on compress coding. The authors present an entire system for MPEG-4 facial animation, develop a method that begins with cylindrical range and re-flectance data acquired by a Cyberware 3D Scanner and construct an efficient and fully functional model of the subject's head from a generic mesh model. The authors combine parameterized mode with muscle model to manipulate the subject's head model and then synthesize facial expression. Using those methods, the authors get a good result.