Abstract: The classical Principal Component Analysis (PCA) for image feature extraction is usually based on vectors, which makes it very time-consuming, and the class information in the training sample has not been utilized fully also. To overcome these two drawbacks of PCA, this paper proposes a novel and efficient PCA method based on original image matri-ces directly. It can extract the discriminant information included in the class mean images. Hence, the proposed method has better discriminant performance than classical PCA. Ex-perimental results on ORL face database show the proposed method is more powerful and efficient than the classical PCA and Fisher linear discriminant analysis.