Abstract: In order to obtain the correct matching pairs of SIFT (Scale Invariant Feature Transform) feature of two images， which seems to be the most appealing descriptor for practical uses according to a recent comprehensive test result on large scale database， this paper proposes a two stages iterative procrustes matching method. Firstly this method finds the initial matching SIFT feature pairs based on SIFT descriptors，there usually give rise to many mis-matching feature pairs in this case. Considering the distribution information of the SIFT features， the first iterative procrustes matching stage is used to discard the mis-matching feature pairs efficiently. Unfortunately some correct matching feature pairs would also be discarded in this stage. In the end， the second iterative procrustes matching stage is used to find back those correct matching feature pairs， which are discarded in the process of the first stage. Experimental results show that the algorithm can not only discard those mis-matching feature pairs existed in the initial matching results efficiently， but also find back those correct matching feature pairs perfectly.