Abstract: Using space-borne GPS observation data of GRACE for kinematic precise point positioning, it is found that the positioning error has obvious periodic characteristics. The first characteristic is that the waveforms of the positioning error sequences at intervals of four days basically coincide, and the second characteristic is that the positioning error sequences of two adjacent days have a tendency to shift along the abscissa. Based on the above two characteristics, two methods for eliminating periodic errors are proposed, which are the periodic item error elimination method based on median modeling and the periodic error elimination method based on translation prediction. The former improves the accuracy of the intermediate sequence in a group of data series obviously, while the latter is more stable on the whole. A comprehensive analysis of the improvement effect of the two methods is carried out. Finally, the positioning accuracy is improved from 5 dm to 2 dm. The RMS values in the three directions are obtained significantly reduced.