Reducing the impact of optical aberrations and random noise on the accuracy of star centroid location extractions is difficult using traditional centroid algorithms. To solve this problem, the star image of a star sensor is analyzed herein and a star sensor centroid localization algorithm based on star image resampling is proposed. The algorithm employs the modulation transfer function of the star sensor optical system and pixel frequency response characteristics of the image sensor. The point spread function of the defocused optical system is calculated based on the Fraunhofer diffraction theory, and the systematic error of the resampling-based centroid algorithm is simulated. Simulation results show that the root mean squares of the systematic errors under different aberration conditions are <0.01 pixel. Results of the error-measurement experiments of centroid extraction system performed on a star sensor show that the systematic error of the resampling-based centroid algorithm is 0.008 pixel, which is 66% lower than that of the traditional sinusoidal curve compensation method. The proposed algorithm has high precision and is not affected by the aberrations of the optical system; therefore, it is an effective method for improving the accuracy of star sensors.