Motion estimation (ME) is a major part of modern video codecs, which requires a huge amount of computations. To overcome this drawback, various techniques have been proposed. In this paper with the aid of fuzzy inference, an efficient algorithm is devised. The proposed algorithm exploits spatial correlation as well as temporal correlation among motion vectors. This algorithm uses fuzzy rules to determine the initial motion vector. After that, a local search around initial vector is carried out. In order to decrease the complexity of the algorithm, a look-up table is used. In this table, defuzzified values are stored. Also, to further reduction of complexity, few computations are performed in the proposed algorithm for stationary and quasi stationary blocks. To determine which block can be regarded as stationary or quasi stationary block, a simple comparison with a predefined threshold is done. The experimental results show that the proposed algorithm performs better than other fast block matching algorithms in terms of picture quality and computational complexity.