The GPS network differential technology, which based on CORS (Continuous Operation Reference Stations), has a wide coverage, high reliability. And now, it has become the representative of the most high-precision level in real-time kinematic positioning. It decreases correlative errors, such as atmosphere delay, satellite orbit errors, through integrating data from continuous operation stations. Then generates the RTCM correction, and broadcasts the information by wireless communication technology. As to VRS (Virtual Reference Station)/RTK technology, the structure of CORS fiducial station networks was analyzed. Traditionally, based on the principles of Delaunay triangulation, it had realized automatic networking between stations, and dynamic updating such as point inserting and deleting, and also the problem of selecting the best reference station in mobile positioning was solved. But, this sort of triangular network also has some shortcomings. Such as, it only adopts one baseline estimation, and redundant observation information has not yet been fully tapped, this would go against large-scale reference station network initialization. Also, it has less independent baseline information, this would affect the interpolation precision of the mathematic model, especially in some areas where have large elevation difference. The third is that, in some areas where outside the basic solution unit, the correction accuracy would have fallen dramatically with the increase of the distance. Aiming at the shortcomings of VRS positioning algorithm based on traditional triangulation network, a new VRS localization algorithm based on star network is proposed. This paper establishes a method of GNSS/CORS baseline solution based on star structure, and presents a method of ambiguity solution, and further research about the ionospheric and tropospheric correction algorithms are analyzed respectively, and two experiments are carried out to verify and analyze the calculation and correction of network ambiguity in the new network and traditional triangulation network. Based on the self-development software, experiments were implemented through actual GNSS data. These experiments show that the proposed algorithm model based on star structure can significantly accelerate the time of ambiguity fixed, up to 45%, and has higher accuracy and reliability in generating corrections. These research results have great theoretical significance for the wide application of GNSS/CORS.