Common mode error is one of the main error sources of GPS time series analysis. This paper extracted the common model error (CME) using Principal Component Analysis (PCA) from the data of 14 continuous GPS reference stations from Jan 1st ,2005 to Dec 31th,2009. The mean baseline of these GPS sites is 1100 km. The contribution of first principal component using PCA method is 51.1%, 52.4% and 41.7%, respectively. The contribution of second principal component is 11.5%, 13.1%, and 10.6%. The contribution of third principal component is 8.8%, 10.8% and 10.0%. While the contribution of the fourth principal component is only 5.2%, 4.4% and 7.4%, the sum of first three principal components is 71.4%, 76.2% and 62.3%, which indicates that most valid information are contained in the first three components. Thus we take the sum of the first three components as CME. After removing the common mode error of the GPS station time series, the average RMS decline of the N, E, and U are 47.2%, 50.5%, 37.7%, respectively.