Gait can be easily acquired at a distance, so it has become a popular biometric especially in intelligent visual surveillance. In gait-based human id en- tification there are many factors that may degrade the performance, an d noise on human contours is a significant one because to extract contours perfe ctly is a hard problem especially in a complex background. The contours extracted from video sequences are often polluted by noise. To improve the performance, we have to reduce the effect of noise. Different from the methods which use dynamic time warping (DTW) in previous work to match sequences in the time domain, a DTW- based contour similarity measure in the spatial domain is proposed to reduce the effect of noise. The experiments on a large gait database show the effectiveness of the proposed method.