Firstly, the key feature points which have much less reaction under changing appearances are captured from a tested face. Then the transforming vectors between the relative coordinates of the captured points and each corresponding point in the face model are used to estimate the pose of the tested face. At last, a synthetical face shape model is established by using the captured points and the estimated pose vector. The shape model would be robust under changing appearances and poses because the points in the model are unchangeable under different appearances. In addition, this paper estimates the pose of the tested face correctly. The shape model will be fit for face recognition.