The key frame extraction is one of the key technologies in Content-based video retrieval( CBVR).In order to extract self-adaptively and effectively the key frames in the video,this text proposes an algorithm of key frame extraction basing on improved three-dimensional ant heap.First, this algorithm extracts three-dimensional feature vectors from color space of H-S-V for every frame and shows them as the points of three-dimensional European Space.Then,we divide all the frames into clusters through the algorithm of improved three-dimensional ant heap to extract the key frames in video.Through simulating the new arithmetic through the MATLAB and comparing it with the traditional arithmetic,results show that the recall ratio and precision ratio are improved in some extent.