Establishing reliable point correspondences between two images is a fundamental problem in computer vision. This work focus on the evaluation of candidate correspondences in matching points based on SIFT features, which is one of the most popular local descriptor of feature points. We introduced a novel concept of ambiguity measure (AM), and proposed several AMs based on study of the typical “distance ratio between the nearest neighbor and the second”. We also studied the AMs theoretically and experimentally. It is shown that the capacity of AM depends on whether the matching degree between the putative correspondence points are included in AM effectively. This research are helpful both for understanding the available method NNDR, and for designing more effective methods.