Abstract: Segmentation is an important part of fingerprint image preprocessing.Based on the segmentation of variance method，which the artificial choice of threshold is difficult and inaccurate，as well as the incorrect segmentation caused by random noise，this paper proposes a local threshold segmentation on the combination of the variance and gradient.Compared with some typical fingerprint segmentation algorithm in the experiment，the local threshold segmentation is efficient and fast，which is suitable for low-quality fingerprint images，especially have better results to the images which is difficult to the global threshold segmentation algorithm.