Abstract: Based on the correlation analysis of position time series for GPS stations, a spatial filtering method, correlation weighted stacking filtering, was put forward to remove common－mode component (CMC). This method uses correlation coefficients as weights in spatial filtering, and takes into account such factors as site distances and thewhole level of correlation. There is no need for the assumption of spatial homogeneity as adopted by current filtering techniques. The method has been used to analyze the position time series of 310 GPS stations in the U S PBO　(Plate Boundary Observatory) network, 33 sites in China and about 200 global GPS stations. The results show that there are obvious CMC of 200 km scale, and the common part of CMC for different stations decreases as the baseline distance increases. Near about 2000 km, the position residual time series of GPS sites are usually no longer correlated. Compared to the traditional regional stacking filtering, the correlation weighted stacking filtering can generally introduce 5%mdash;20% RMS (root mean square) residual improvement for small scale GPS networks. It can also extract CMC of individual spatial scale and separate single site anomalies, increasing our ability in detecting weak tectonic signals and searching for CMC origins.