Abstract: GPS (Global Positioning System) time series spatial filtering can improve the signal-to-noise ratio of the observations, which is beneficial to obtain more accurate deformation information. The spatial filtering result by using stacking filtering method is not unique, as the number of stations and the spatial scale are different. This is not good to study the deformation information in GPS time series. In order to weaken the influence of spatial scale on spatial filtering method, an improved stacking filtering algorithm is proposed. This method introduces correlation coefficient factor and distance factor as the weights, and it does not constraint by spatial scale. This method is applicable for the regions with different GPS station distribution. We study the spatial filtering method for 260 GPS continuous observing stations in China, and the observation time series is from 2010 to 2017. The results show that:(1) Compared with the correlation weighted stacking filtering algorithm, the method is more effective to extract the common-mode errors from the GPS time series. The stacking filtering method are less constraint by the spatial scale, which take the correlation coefficients between GPS stations and distance factors as weighted. (2) Compared three kinds of stacking filtering methods with different distance factors as weighted, the method with the inverse distance factor obtained the better spatial filtering result. (3) The residuals of GPS time series can be reduced by 30%-40% and the precision of GPS velocity field increased by 30%-40% after spatial filtering by this method. This spatial filtering method can obtain better GPS deformation field estimation. And it also provides a reliable data foundation for studying the crustal movement and its dynamic mechanism in China.