Abstract: The theory and algorithm of determining system errors of CHAMP accelerometer are presented based on energy conservation method. The time series of scale factor, bias and bias drift is analyzed using about two and a half years' accelerometer data of CHAMP from August 2001 to December 2003. The long-term and principal periods variations are obtained by using LS spectral analysis as well as the robust one. The results show that there are clear periodic variations in the time series of scale factor, bias and bias drift. For scale factor, the principal periods are 441 d, 220 d, 126 d and 13.4 d. 441 d and 220 d periods make great contributions to scale factor. For bias, the principal periods are 220 d, 126 d and 13.4 d, and they make almost the same contributions. For bias drift, the principal periods are 220 d, 13.4 d and 9d, and 13.4 d makes the greatest contribution. There are obvious long-term variations in the time series of scale factor and bias except bias drift. It is shown that the results of the robust spectral analysis are superior to that of LS one in the whole. Further investigations about why there are long-term and period variations in system errors of CHAMP accelerometer are necessary.