Joint source-channel decoding (JSCD) techniques have been proposed to reduce symbol error rate and bit error rate in decoding a sequence through exploiting the residual redundancy, but these techniques always employ only spatial residual redundancy. In this paper, a new JSCD scheme which exploits both spatial and temporal redundancy in motion vector data (MVD) is proposed. Simulation results show that our proposed scheme leads to obvious improvement over the scheme without using the a priori information of temporal redundancy.