Abstract: This study takes typical mountainous towns in different basins in Yunnan Province as the analysis area. By removing and weakening the effects of altitude and latitude on the multi-year mean temperature field, in the temperature field of typical mountain towns at the same altitude and latitude, based on physical geographic and socioeconomic data, the global spatial regression model of influencing factors and temperature distribution was established by the ordinary least square(OLS), and the local spatial regression model was established by geographical weighted regression(GWR). Result shows that the main factors that had significant influence on the air temperature in mountain towns were GDP, topographic uplift height, basin opening angle, the third industry ratio and vegetation index. The R² of the ordinary least square adjusted with the air temperature is 0.789, the local spatial regression model established by geographical weighted regression is more reliable. Taking reserved mountain towns as test samples, the independence test of the temperature model was carried out. The plan of eliminating altitude and latitude MAE is 1.09 ℃, and the RMSE is 1.35 ℃, the plan of eliminating altitude and latitude has better accuracy error to temperature than the plan of eliminating altitude and latitude only and preserving altitude and latitude only.The study reflects that the natural geographical pattern of mountain towns and human social activities play an important role in the distribution and spatial simulation of temperature field.