In the motion estimation of MPEG and H.263,full-search algorithm doeswell in matching accuracy, but involves in a very high computational complexity. Some fast search algorithms such as three-step search still need to be improved in both computational complexity and matching accuracy. The proposed new algorithm, ASA, takes good advantage of both the center-biased motion vector distribution characteristics of the real world video sequence and the correlation among the motions of adjacent blocks. It performs as accurate as full-search, but needs much less computation. Compared with existing algorithms, it is more efficient, more effective and more robust.