In video communication based consumer device applications, the compressed video is extremely fragile to transmission errors due to channel noise. Error concealment (EC) is an efficient way to recover a damaged video sequence. The existing EC algorithms for compressed video typically apply uniform EC mode to entire video sequence regardless of frame features, or lack in sensible mode-switching considerations. This paper proposes a new dynamic multi-mode switching (DMS) EC algorithm. The proposed algorithm provides several EC modes for intra- and inter- frames. The proposed smart decoder is able to adaptively apply these modes in different situations. The mode switching in DMS depends on the spatial/temporal correlation, estimated edge features of the corrupted Macroblock (MB), the smoothness of the reconstructed MB and etc. The proposed DMS algorithm has been evaluated and compared with 3 classical EC algorithms. The experimental results show that the DMS algorithm can achieve significant gains in peak signal noise ratio (PSNR) and therefore improves the decoded picture quality in video applications.