The distance ratio threshold is used for matching in scale invariant feature transform algorithm, which presently is a fixed value in many researches and weakens the algorithm's robustness. The distance ratio threshold was analyzed in detail, and an adaptive method was put forward. In addition, the ratio of random sample consensus optimized points and matching points was used as an approximation of repeatability to identifying the optimum value. The experimental results show that this method can find the optimum threshold value through few iterations, which can improve the matching accuracy while not bring too much calculation.