A method to improve the accuracy of star locations in star images is presented based on image processing.The reasons for generating noises in star locations are researched, and a frequency spectrum of a star location before denoising is analyzed.Based on the analyzed results, a Finite Impulse Response (FIR) filter is designed and realized based on the Parks-McClellan algorithm.Finally, the spectra of the star locations obtained from the star images are analyzed with the FIR filter,and the results show that the FIR filter could eliminate the noises of star locations well.The algorithm is successfully demonstrated by real space experiment data captured by a star sensor in a satellite.The experimental results show that the accuracy of attitude has improved to 7.823″ from 21.921″.