One of important objectives of astronomical observations is to obtain physical information, including positions of measured stars with high accuracy from digitized CCD images. By means of CCD images taken by the 1m telescope at Yunnan Observatory, re-search on the comparison of the accuracy among three often-used centering algorithms is carried out, including Modified Moment, Two-dimensional Gaussian Fit and Median. The positions of 70 stars whose magnitudes vary from 10.2 to 17.6 in 18 CCD images are mea-sured in our experiments. Based on the experimental results, these algorithms are compared and discussed. A visualization, which is a distinct feature of this paper, is used to analyze and select different thresholds. Since each algorithm is programmed using several thresholds, the influence of threshold settings on precision in both x and y direction is also discussed in details. Through a comparison of the same algorithm with different thresholds to choose an optimal threshold as well as a comparison among different algorithms aimed to study the accuracy of centering algorithms, significant finding is displayed. Experimental results show that two-dimensional Gauss method with no threshold fits a stellar distribution best and therefore is a relatively high-accuracy centering method for faint stars as well as bright ones.