A novel video watermarking algorithm is proposed based on data statistical classification of motion vectors.In this algorithm,watermark information relates closely with data characteristic,it doesn't have the physical partitionablity.Therefore,the algorithm avoids the sensitivity to watermark synchronization compared with traditional algorithms.The experimental results demonstrate the algorithm has great robustness and unperceptiveness and can efficiently against temporal synchronization attacks and content-based attacks.When lost frame ratio is lower than 18%,watermark accuracy is higher than 80%.