Abstract: The accurate acquisition of weighted mean temperature is critical to GPS inversion of atmospheric precipitation. Based on the linear regression theory, the correlation between weighted mean temperature (Tm) and surface temperature (Ts) is analyzed. Then linear regression is made to get the empirical formula of Tm with Hong Kong radiosonde data. The results show that the root mean square of the Tm using this paper’s model is 2.356  K, which is 41.94％ higher than that of the Bevis model, and the effect of seasonal variation on the Tm accuracy is not significant. Finally, the model of this paper is applied to remote sensing atmospheric water vapor. The root mean square is 1.807 mm, and the mean deviation is 1.362 mm. The results indicated that the model of this paper for Tm can meet the accuracy of GPS inversion of atmospheric precipitation, and better than the Bevis model.