An improved method of motion fuzzy facial expression feature extraction and recognition is proposed to solve the problem of blur and unclear contour of facial image caused by horizontal motion. Traditional method of time frequency estimation is used to calculate the horizontal motion blur length. Anisotropic factor is redefined in anisotropic Gauss filter to form an elliptical Gauss window.The new filter is introduced into the Fourier transform of rotation invariant local phase quantization（RILPQ） algorithm for feature extraction.The experimental result based on JAFFE dataset shows that the proposed method not only can more clearly extract facial feature for motion blurred facial images

but also has higher recognition rate compared to the RILPQ algorithm.