This paper proposes an automatic 3D reconstruction approach for a Chinese talking face by a generic model and a single image. Firstly, an improved color-based ASM method is used to detect the face area and get the 2D face feature points automatically from the given image, which is not restricted to full frontal one. Then, color information is used to correct the location of face feature points. Finally, after text mapping, a particular and realistic 3D face model is deformed from a generic model. Using ASM face feature points extraction and correction based on skin color model, the problem of side face information missing is successfully resolved. Depending on only one image and one generic model, the computing cost of memory and time is largely reduced. The 3D face reconstructed can be easily deformed to form different expressions and mouth shapes. Experiments show that this approach is fast and efficient and has an output of a lifelike Chinese talking face.