Abstract In high-precision GPS analysis, the surface temperature and pressure of GPS sites are used to derive a priori zenith tropospheric delay. By comparing the different estimates of site position obtained using the global model of pressure and temperature GPT and GPT2 respectively, we find that the deviation of a priori zenith tropospheric delay due to the site pressure deviation cannot be absorbed completely by the delay parameters. Further, it will cause the deviations of site position estimates, especially for the vertical component; these deviations also depend on site latitude and elevation-dependent data weighting used in the data analysis. The site temperature deviations will not affect the site position estimates clearly. However, statistical analysis based on a large amounts of data show that the accuracy of position estimates decreases with the increase of site temperature. Taking the results of regional network in 2013 as an example, when site temperature increases from 0 ℃ to 30 ℃, the repeatability of daily solutions occupied 4 d decreases from about 1 mm to about 2 mm, and from about 3 mm to about 6 mm for the horizontal and vertical components, respectively.