In frequency source designed by using GPS satellite signals to tame the ground crystal oscillator, there is interference noise in frequency deviation between GPS and crystal oscillator.If using Kalman filter to process the frequency deviation directly,outliers in the signal noise will influence the accuracy of filtering.In this paper,a Kalman filter algorithm based on minimum sum function estimation was proposed,which used the weighing matrix derived by influence function of M-estimation to judge whether the frequency deviation has outlier,thus to correct it for new filter information.The method has been used for some frequency calibration systems.The results show that the influence of outliers on the filtering accuracy can be effectively suppressed,and the output frequency accuracy is higher than 5.0×10^-12.