In this paper we propose a new model of an upper body of human for visual animation of Sign language, and a description method of human motion based on the model. The gestures of Sign language are represented by shapes and their changes of hands and fingers, and motion of arms. The shaping of hands is described as a code of alphabets and numbers, where alphabets denote shaping and each finger is numbered. While a number of joint angles in a time sequence are adopted in former studies, a few 3D position parameters proposed in this paper play roles as key frames in an animation, and can generate sufficiently smooth trajectories by interpolating with a spline function between them including inverse kinematics.