Abstract: Based on the complex topographical features and sparse uneven observation sites in Sichuan Province, terrain factors (slope and aspect) and vegetation index were introduced in this paper. The method of (mixed) geographically weighted regression kriging ((m)GWRK) which took into account the non-stationary of spatial relationship was adopted to study the interpolation method of monthly mean temperature and the precision analysis of the estimation results. In different seasons and different regions, the estimation results of (m)GWRK and regression Kriging (RK) based on global regression were compared. The results indicate that the coefficient of determination(R²) of regression relationship of RK, GWRK and mGWRK are 0.795, 0.922 and 0.911, respectively, and root meansquare error of these three methods are 0.83℃, 0.64℃, 0.55℃, respectively. This implies (m)GWRK is better than RK in ability to interpret the target variable and estimation accuracy. Compared with RK, the improvements of (m)GWRK on estimating monthly average temperature have the characteristics of seasonal and regional differences. The improvement is more significant in winter half year than in summer half year. And in northwest and southwest Sichuan, where topography changes acutely, the improvement is greater than in basin where topography changes gently.