Abstract: An improved fast subpixel search algorithm is proposed based on the characteristic of the matching error function with the single peak error surface. Firstly, a best halfpixel position is approximated by a five point error surface model based on the results of the integerpixel motion estimation, and then the search range and search template are determined for the halfpixel. Secondly, a search method of the quarterpixel is selected according to the halfpixel search results to complete the subpixel search. The experimental results show that the proposed algorithm can reduce nearly 60% of the calculation in the subpixel search with averagely a 0.043 dB decrease of peak signal to noise ratio, which improves the coding efficiency while guaranteeing the subjective and objective quality of the reconstructed image.