In order to improve the range-measuring accuracy in stereo vision,a sub-pixel parallax is needed. The Mutual Information（ MI） theory is combined with multi-resolution method to realize the goal of sub-pixel point matching. Firstly,Bouguet algorithm is used to rectify the left and right images making their epipolar line forward-paralled. Then Harris corner detector is brought to find a most characteristic corner as candidate matching point. After that,Most Cross Correlation matching rule is introduced to search the matching point. The 20 × 20 areas whose center is the left and right matching point are magnified by 10 and100 times respectively. The low-resolution image is registered with MI theory,followed by the seeking of higher precision in the high-resolution image. Finally,combined with the integer-grade parallax,we can get the sub-pixel parallax. The experimental result shows that the method used in this article can improve the precision to 0. 01 pixel level.