The precision of centroid location for star images is an important factor, which affects the correction of star identification and the precision of star attitude determination.In this paper, the error model using the traditional Center of Mass(COM) is analyzed and it is shown that there are systematic errors and random errors in the COM.For the systematic errors, this paper firstly analyzes the relation between the systematic errors and the calculated centroid position by a simulation, and then adopts the least square curve fitting algorithm to estimate parameters of bivariate polynomial and to compensate systematic errors.For random errors,it presents an edge threshoold method and a star image elimination method to remove errors.Furthermore,the CCD imaging model is analyzed and a Bivariate Polynomial Compensation Template(BPCT) is proposed.As the star intensity distribution is close to the Gaussian point spread function model and the BPCT is related to the dispersion size of star intensity distribution, the different BPCTs are calculated to compensate the systematic error of COM according to different sizes of star dispersion.The simulation results show that by using the BPCT, the accuracy of the star position can be enhanced from 1/20 pixel to 1/200 pixel,which meets the requrirements of star trackers for the high accuracy centroid location.