A novel feature points based gait authentication method is introduced in this paper, which uses the acceleration signals acquired from an ankle-mounted 3-axis accelerometer. Feature points are extracted from original gait samples. A dynamic time wrapping algorithm is employed to match the feature points of different samples and calculate the distortions. Based on these distortions, a multi-criterion model is designed for authentication. The experimental result shows that the extracted feature points can represent the original signals good enough in authenticating with one accelerometer, and the equal error rate of this method is only 3.27%, better than that of the previous literatures reported.