Abstract：This paper proposes a high accuracy prediction method for Earth polar motion parameters, using a dual differential least-squares (LS) and autoregressive (AR) model. Firstly, Earth polar motion parameters are calculated by dual differential methods to obtain differential results. The stationarity of differential polar motion parameters is improved. The LS method is utilized to fit the differential polar motion parameters to obtain the residual results. Secondly, the AR model is utilized to analyze the residual polar motion parameters. Thirdly, the extrapolated results of LS and AR model prediction results are combined to obtain the differential prediction results. Finally, the differential prediction results are calculated by inverse dual differential methods, to obtain the final prediction results. The proposed method is applied to the real Earth polar motion parameters prediction. The results show that 1 day’s PMX prediction accuracy is better than 0.25 mas, and 1 day’s PMY prediction accuracy is better than 0.2 mas. These results are compared with EOP_PCC results. It shows that this paper’s polar motion parameters short term prediction accuracy is close to EOP_PCC good prediction accuracy, and that 1 day’s prediction accuracy is a little better than EOP_PCC 1 day’s prediction accuracy.