The proper characterization and suppression of GNSS errors is critical in maximizing the accuracy achievable for regional GNSS networks. Common Mode Error (CME) is one of the most dominant errors in daily GNSS solutions, and Principal component analysis (PCA) is commonly used to extract. PCA is a multivariate statistical analysis approach that uses a linear transformation of multiple variables to select fewer important variables. Orthogonal transformation transforms its component-dependent original random vectors into a new component-independent random vector. This method aims to use the idea of dimension reduction to convert multiple indicators into a few comprehensive indicators. In other words, the composite indicator provides most of the information for the original indicator. This feature extraction method has been widely used in geophysical signal processing, geographic information systems and remote sensing. However, how to define and quantify the CME is still an open question.The usage of PCA approach has one condition that the area is evenly distributed, which limit is application. In order to improve it, this paper introduced the Geographically Weighted Principal Component Analysis (GWPCA) method. After adjusting the influence of the spatial effect, the accuracy of the CME has been greatly improved. During the application of GWPCA, one of the key problem is to chose the proper bandwidth,with the change of bandwidth, the information of each geographical location will change accordingly. The larger the bandwidth, the wider the included area and its results are more closer the PCA method. The smaller the bandwidth, the factors that affect the CME around the study point will become the main factor, and the weight of the peripheral influence will increase. Different bandwidth, the weight of each influencing factor is not the same, appropriate and reasonable selection of bandwidth using GWPCA method can improve the accuracy of CME and better reflect the spatial heterogeneity.