Abstract: Using the statistical synthetic regression algorithm by the USA, the temperature and moisture profile of the atmosphere is retrieved based on the MODIS data, which, as an initial condition, is then employed for estimating the atmosphere transmittance through Pressure-Layer Fast Algorithm for Atmospheric Transmittances(PFAAST), and thus the atmosphere temperature is retrieved via the nonlinear physical retrieval algorithm. The results show that the atmospheric temperature error is within 2 K at and above the top of boundary layer with a larger error within boundary layer, that is positively correlated with the aerosol optical depth (AOD) increment as well as the estimated error of surface temperature, but with poor correlation with atmospheric moisture mixing ratio. On the basis of the radiative transfer theory and sensible experiments, the effect of AOD and surface temperature on retrieval error is analyzed with the result that the surface temperature weighting to different spectral band is enhanced with increasing surface retrieval error, suggesting that a reasonable surface temperature precision is helpful for improving surface temperature weighting. Aerosol concentration is high within the boundary layer over the desert and the Gobi region in the northwestern China, which reduces the atmospheric transmittance therein. Besides, the surface temperature weighting and air temperature weighting to infrared band should be decreased in case the infrared remote sensing data are used. In view of that the sand aerosol effect within boundary layer is omitted, the atmosphere transmittance retrieved is overestimated and thus the surface temperature weighting and air temperature weighting are overestimated as well, causing that the air temperature retrieved is underestimated. According to the observed aerosol optical depth from sunphotometer, an improvement in the AOD effect and thus the surface temperature is obtained resulting from improved temperature profile that is able to reflect the real atmosphere structure within boundary layer.