This paper presents a new technique called edge detection for fast block-matching motion estimation. In order to find the matching macroblock for the current macroblock from previous frame, this technique classifies the current macroblock into shade and edge. The shade macroblock has a high probability to move in the same direction as its neighbouring macroblocks. This property has been used to decrease the computations of Mean Predictive Block Matching algorithm. The proposed technique used only the motion vectors of the neighbouring macroblocks and ignore other motion vectors that were utilized in the first search step of Mean Predictive Block Matching algorithm. Experimental results of various video sequences types showed that the proposed technique reduces the average number of search points required per macroblock for the videos, as well as keeps or enhances the resolution of some decompressed videos in comparison to the Mean Predictive Block Matching algorithm and the standard block matching algorithms.