High precision of the prediction of searching initial point can reduce the number of search points to improve the search speed and search precision. The median, the mean, the left block, the SAD (Sum of Absolute Difference) comparison and other methods of the prediction of searching initial point were theoretically and experimentally analysised. And according to The correlation of adjacently blocks’ motion vectors and the characteristic of the center-biased characteristic of motion vectors in image sequences, a novel method of the prediction of searching initial point was proposed. The novel method assigns different weights to the adjacent blocks. Experimental results show that the novel method gains better reduction(29.24) in the number of searching point over the different kinds of criterion testing video sequence and impoves 1.71dB in peak signal to noise Ratio(PSNR).