Abstract: The weighted average temperature is an important parameter in the research of atmospheric precipitable water vapor, which directly affects its precision. In view of the less weighted average temperature model suitable for Gansu Province and the low accuracy of existing models, we mainly use the data of 7 atmospheric sounding stations in Gansu from 2013 to 2017, and statistically analyze the spatial and temporal characteristics of the weighted average temperature in Gansu Province. Then based on the principle of least squares, localized partition, sub-quarter and sub-month models are established, and the accuracy of the model is verified by the existing models. The results show that the weighted average temperature in Gansu has obvious spatial and temporal characteristics. Therefore, it is very necessary to establish partition, sub-quarter and sub-month models. The precision of partition, sub-quarter and sub-month models is better than the existing models, which has important practical value.