SIFT (Scale invariant feature transform) descriptor is one of the most effective local features that is used for scale, rotation and illumination invariant of automatic image registration. In this paper a fast bi-directional matching algorithm based on SIFT was proposed for improving match accuracy and reducing match time. Firstly the single-feature points and multi-feature points in two images were extracted, and then matched all these feature points respectively by using the BBF (Best Bin First)-based bi-directional matching algorithm. The integrated matching correspondences were the final matches. The experimental results showed that the proposed algorithm can reduce mismatch probability and decrease the matching time.