Abstarct: Block matching motion estimation is a key component in video compression. Motion estimation (ME) is a core part in video coding, which can effectively remove temporal redundancies among the adjacent video frxames to achieve high coding efficiency. In this thesis, we propose two block matching algorithms to reduce the matching block searching time and improve the quality of the reconstructed video. The algorithms consist of two stages: the initial search and the local search. In the first algorithm, an adaptive rood pattern with unequal arms and position of left and above neighboring motion vectors are used in the initial search; and the 3×3 square pattern is used in the local search. In the second algorithm, we have proposed three different patterns in the initial search. First, a large square pattern, in which the horizontal and vertical step sizes are determined according to horizontal and vertical components of the left neighboring motion vector. Second, the pattern of four points of the large square pattern, which are located in the left neighboring motion vector range; and then complete the search pattern baxsed on error values obtained from these points. Third, the pattern of four points of the large square pattern without considering of matching error value of these points, which are located in the left neighboring motion vector range. The 3×3 square pattern is used in the local search. The results of the proposed algorithms for 9 test video sequences and comparison with standard block matching algorithms shows that the block matching searching time is reduced and algorithms are about 2~12 times faster than that of full search algorithm and the PSNR, especially for video sequence with large motion is improved.