This paper presents a new point matching method to solve the dense point-to-point alignment of scanned 3D faces. Texture maps of 3D models are generated at first by unwrapping 3D faces to 2D space. Then, we build planar templates based on the mean shape computed by a group of annotated texture map. 34 landmarks on the unwrapped texture images are located by AAM and the final correspondence is built according to the templates. Comparing to the traditional algorithms, the presented point matching method can achieve good matching accuracy and stability.