In this paper, we proposed a partitioned linear minimum mean square error estimator (P-LMMSE) for error concealment. The proposed P-LMMSE estimator adopts the multi-hypothesis motion compensation (MHMC) technique to reconstruct the corrupted block, in which the lost blocks are predicted by a linear combination of motion compensated blocks (hypotheses). In our proposed P-LMMSE estimator, the weighting coefficients are optimal in the sense that they minimize the mean square error. In addition, our proposed estimator exploits the properties of the hypotheses to improved accuracy of prediction. Each hypothesis has its own assumption and hence works well only in a particular situation, or equivalently, the statistics in different situations are not the same. Therefore, the dataset is divided into finer partitions and the weighting coefficient set in the most appropriate partition is selected to reconstruct the corrupted block.