In this paper, we propose an efficient motion vector recovery algorithm for the new coding standard H.264, which is based on a polynomial model. To achieve better coding efficiency, the motion estimation scheme used in H.264 is different from previous coding standards. In H.264, a 16/spl times/16 macroblock can be divided into different block shapes for motion estimation. Each macroblock contains more motion vectors than previous coding standards. For nature video, the blocks within a small area likely belong to the same object, hence the motion vectors of neighboring blocks are highly correlated. Based on the correlation of neighboring motion vectors, we can use the motion vectors that are adjacent to the lost motion vectors to constitute a polynomial model, which can describe the change tendency of motion vectors within a small area. Through this model, the lost motion vectors can be predicted and the lost macroblocks can be reconstructed. Different video sequences are used to test the performance of proposed method. The simulation results show that the quality of corrupted video can be obviously improved by proposed algorithm.