Abstract: In order to quickly extract the facial feature points, we propose a fast eye center and corner localization algorithm which is based on the reconstructed gradient feature. Studies have shown that gradient feature of human eyes is radiating distribution centered on pupil. Based on that, the algorithm proposed in this paper compute the center to detect the location of the pupil and the eye corners. Then necessary adjustments with respect to the relative location of both pupil and eye corner is employed to eliminate the gross error. Only the original gradient information is used which is invariant to affine lighting changes. The images from both Muct and Bio ID database as well as low-resolution web camera are used to test the proposed algorithm. The experimental result demonstrate that the proposed algorithm has high detection precision with low computational cost, which can meet the demand of real-time applications.