Abstract: In order to improve the centroid sensing accuracy of spot image of Shack-Hartmann wavefront sensor (S-H WFS), the error sources of centroid detection were discussed in the paper. The influence of detection window, threshold selection and centroid algorithm on centroid detection accuracy was analyzed. On this basis, detecting window selection method based on template match and adaptive threshold selection method based on single spot were proposed. The methods proposed in this paper were combined with the center of weight algorithm for improving the centroid detection accuracy. The accuracy with the comprehensive method increases 40% comparing with traditional method when signal-to-noise ratio(SNR) of the spot image of S-H WFS is more than 3. The methods was applied to a S-H WFS working in visible spectrum, the results showed that maximum centroid departure decreased from 0.83 pixel to 0.15 pixel. The work in this paper has a high value of practical application.