As the S-curve errors are important parts of centroid location errors for star sensors, this paper explored the sources of S-curve errors combining with the physical process of centroid location and a simulation. The specific effect of each error source was analyzed and an analytical expression of S-curve errors was calculated by a frequency domain method. The experiments using a star sensor were performed and the S-curve error in the center of the Field of View (FOV) was collected and was compensated using a sine model. The compensation effects on the S-curve errors in the whole FOV were analyzed by the same compensation model and the calibration data were also compensated. Experimental results show that the standard deviation of S-curve error is 0.048 pixels in the center of the FOV, and 0.027 pixels after compensation, therefore the precision of centroid location is improved by 43.8%.Furthermore, after compensating with the same sine model in the center of the FOV for whole field-curve errors, the precision of centroid location in the whole FOV is improved by 35.7% at least and the precision of calibration is improved by 31.7%. It concludes that the S-curve errors are important errors of star sensors and they can be significantly compensated by using the sine model.