Abstract: Weighted average temperature (T_m) is a very important conversion parameter in atmospheric water vapor retrieval by GNSS. Based on the measured data of Baise, Guilin, Wuzhou and Nanning sounding stations in Guangxi for 15 years (2003-2017), the correlation between T_m and surface temperature T_s, water vapor pressure e_s, latitude φ and elevation H is analyzed with the T_m calculated by numerical integration method as the reference value, and the weighted average model GXT_m in Guangxi is established based on the least square principle. The results show that: T_m is positively correlated with surface temperature T_s, water vapor pressure e_s, and negatively correlated with latitude φ and elevation H; the overall prediction effect of weighted average temperature model GXT_m is better than XIE Shaofeng Guangxi model, YAO Yibin model and Bevis global model, and its accuracy is 8.4% higher than XIE Shaofeng Guangxi model, 25.7% higher than YAO Yibin model, and 28.4% higher than Bevis global model.