A novel image synthesis method of human face with various expressions is presented. In this method the geometrical and textural information implied in expressions are uncoupled. A 2D grid is used to extract the geometric variation of human face with different expressions. A Principal Component Analysis （PCA） is used so that the grid Sequences can be expressed by a few eigengrids. Therefore a parametric model can be developed to control the expressions effectively. During the image synthesis process

image warping and morphing techniques are employed to combine the textural information

supplied by a few reference image frames with the geometrical information

represented in the grid model

so that vivid facial images with different expressions can be preduced. The method presented in this paper has been investigated with some preliminary experiments. Its effectiveness has been proved.