Abstract: Objectives: Multipath error is one of the main error sources in high-precision global navigation satellite system (GNSS) data processing. Non line of sight (NLOS) signal and multipath have different characteristics, which may lead to significant errors in centimeter to millimeter GNSS positioning. However, the current multipath error mitigation methods for carrier phase observations do not effectively distinguish them. Methods: In this paper, the NLOS signals of carrier phase observations at the static station are detected by using 3D point cloud data, and the correction effect of multi-point hemispherical grid model (MHGM) on multipath and NLOS is evaluated. Based on the original MHGM, an improved multipath error modeling strategy to eliminate NLOS signals is further proposed. Results: In the experiments, the statistics of double-differenced residuals in the ambiguity-fixed periods show that MHGM can mitigate the multipath and NLOS errors by 73.5% and 81.2% respectively, but the accuracy of NLOS affected observations after MHGM correction is still significantly lower than that of multipath observations. After the NLOS signal is eliminated in the multipath error modeling and application stage, the RMS of double-differenced residuals in the ambiguity-fixed periods is further reduced, and it is increased by 8.8% compared with the absence of NLOS. Kinematic positioning results show that under the condition of multiple systems and sufficient number of available satellites, MHGM can improve the three-dimensional accuracy of positioning results by 68.6%, while using MHGM with NLOS signals eliminated, the improvement rate can reach 76.0%. Conclusions: It shows that MHGM with NLOS elimination has better correction effect on multipath error, and can achieve a higher positioning accuracy in complex environment.