Abstract: Robustness to collusion attack is one of the most challenging issues in video watermarking. The motion compensated temporal frame averaging (MCTFA) attack removes the temporal uncorrelated watermark by exploiting the temporal redundancy among consecutive frames. In order to address this problem, a motion coherent video watermark, which changes according to the motion in the video frames, is needed. In this paper, we propose a novel watermarking scheme that can effectively combat the MCTFA attack. The method relies on the macroblockbased motion estimation, which conforms to the MPEG compression standards. A motion coherent watermark is generated according to the obtained motion vectors, and embedded into the video frame in the spatial domain. We also prove that embedding motion coherent watermark will not affect the motion vectors of the video, which make it possible to extract the embedded watermark from a watermarked video in the detector, without the need of the original video. Normalized correlation is then calculated to determine whether the received video contains a particular watermark. Experiments show the effectiveness of our method. Besides, the motion coherency of the watermark is further verified by the method provided by P.Vinod et al［5］.