Summary: Motion information is special to video. It provides enough space for watermark embedding. However, the watermark embedded in motion information is usually not robust. In this paper, a robust watermarking algorithm which embeds watermark in motion vectors of P frames is proposed. The reason why the watermark embedded in motion vector is not robust to recompression is that the watermarked motion vector will be changed when the video is recompressed. In order to improve the watermark robustness to recompression, we present to modify the pixel of the current block while embedding watermark bit in its motion vector. In order to minimize the perceptual degradation of video quality, we adopt our previously proposed geometrical method. The experimental results demonstrate the strong robustness of the scheme against common signal processing operations.