This paper proposes a gait recognition algorithm based on three kinds of features from the image's silhouette. It extracts silhouette features by establishing dynamic information of invariant moment and percentage of frame difference, combining the modified angle distance of static information. It realizes the gait recognition via a modified K neighbor method. The experimental result shows that the recognition ratio achieves 91.94% with the static featureis only, and reaches 99.19% with the dynamic features under the improved classifier.