The temperature difference between the surface and the air (dTsa) directly indicates the intensity and heat fluxes of land-atmosphere interaction. Considering the effects of surface characteristics and air condition on the surface temperature, using 1 km data from the MOD02 thermal infrared bands of the EOS/moderate-resolution imaging spectroradiometer (MODIS) on satellite Aqua, other MODIS products and temperatures observed from weather stations at 14:00 China standard time (CST), the study analyzes the relationships between dTsa and brightness temperature in the infrared atmospheric window band (Bt31 and Bt32), the water vapor band (Bt28), the atmospheric temperature band (Bt25), and the CO 2 band (Bt34). A model estimating dTsa is built. The model coefficients are estimated for 96 stations representing 96 sets of surface and atmospheric conditions, and 71 sets of coefficients among them passing 90% confidence levels of estimating dTsa are selected as references. Combined with the probabilistic neural network (PNN) method and nine parameters reflecting surface characteristics in one season and month, the Tibetan Plateau surface is classified as 71 types with 71 sets of coefficients. PNN is certified an efficient classification method for multiple parameters and mass data. Based on PNN and estimated model, estimated dTsa shows 1.36°C root-mean-square error and a standard deviation of 0.74°C, and dTsa distribution exhibits all centers with peak value and valley value of European Centre for Medium-Range Weather Forecasts, MYD07, and simple regression model results, showing its superiority. The model is worthy of further exploration and application in an effort to advance the retrieval of surface energy fluxes from remote sensing.