The variance of the position of point object because of photon noise for centroid estimation algorithm is derived for the first time when the background cannot be eliminated by threshold. On this foundation, the accuracy of centroid estimation algorithm is analyzed by comparing its estimated efficiency (the ratio of the Cramer-Rao lower bound of point object and the variance for centroid algorithm) based on the classic estimation theory. The results indicated that the following factors influence the accuracy of centroid estimation algorithm: SBR (the ratio of the outputs from the signal light and the background), the Gaussian width of signal spot and the size of detected window. The effects of these factors are also described. Crown Copyright (C) 2009 Published by Elsevier B.V. All rights reserved.