Due to channel noise and congestion, video data packets can be lost during transmission in error-prone networks, which severely affects the quality of received video sequences. The conventional inter-frame error concealment (EC) methods estimate a motion vector (MV) for a corrupted block or reconstruct the corrupted pixel values using spatial and temporal weighted interpolation, which may result in boundary discontinuity and blurring artifacts of the reconstructed region. In this paper, we reconstruct corrupted macroblock (MB) by predicting sub-partitions and synthesizing the corrupted MB to reduce boundary discontinuity and avoid blurring artifacts. First, we select the optimal MV for each neighboring boundary using minimum side match distortion from a candidate MV set, and then we calculate the optimal cut path between the overlapping regions to synthesize the corrupted MB. The simulation results show that our proposed method is able to achieve significantly higher PSNR as well as better visual quality than using the H.264/AVC reference software.