Atmospheric water vapor is one of the key factors affecting the retrieval of land surface temperature (LST) from the single-band infrared image. It is conductive to the rational usage of single-band infrared data to find out the influence of spatial heterogeneity of AWV on the retrieval of LST. With the MODIS MOD05_L2 products as data sources, this paper discussed the spatial heterogeneity of AWV at the regional scale and its impact on the retrieval of LST from Landsat image. Firstly, an error model was deduced with regard to the influence of AWV on the retrieval of LST from remote sensing data. Then, in order to explore the retrieval error of LST possibly caused by the spatial heterogeneity of AWV at the regional scale, the authors conducted simulation tests followed by comparative analyses under spatially different distribution conditions of AWV in several regions, particularly for the thermal infrared band of Landsat 5. Results obtained in this paper indicate that the retrieval error of LST caused by the spatial heterogeneity of AWV has to do with the value and the heterogeneity degree of AWV. For example, when AWV had relatively high heterogeneity, an error was relatively great, ranging from 0.5℃ to 2℃ with relative error from 2% to 20% accordingly; by contrast, when AWV was relatively uniform, error was comparatively insignificant and ranged from 0.1℃ to 0.2℃ with relative error from 0.5% to 2% accordingly. In conclusion, with regard to regions with a relatively high degree of atmospheric heterogeneity, it is vital to get atmospheric parameters properly at the regional scale so as to retrieve LST precisely from remote sensing data.