In this study, an atmospheric correction algorithm is designed for Landsat/TM imagery. The lookup tables with multiple scattering and polarization correction are used to remove the reflectance resulting from Rayleigh scattering. The Landsat/TM imageries collected on October 28, 2003, in Taihu Lake, water-leaving reflectance measured by synchronized experiments and the aerosol lookup table are used to estimate the aerosol optical thickness (AOT) and atmospheric diffuse transmittance from Landsat/TM imagery at 15 experimental stations. The inverse distance spatial interpolation algorithm (IDSIA) is used to improve the uncertainty produced by the non-homogeneous distribution of AOT. According to the study results carried out by this paper, it is found that using IDSIA to improve the spatial changes of AOT at least decreases 4.5% uncertainty at TM3 and 16.4% uncertainty at TM4 from non-homogeneous distribution of AOT. The improved performance of IDSIA is fairly obviously. Additionally, water-leaving reflectance of Landsat imageries is estimated by this atmospheric correction algorithm. The stability and accuracy validation results show that the estimation accuracy of water-leaving reflectance is 8.31% at TM1 and 9.56% at TM2, respectively.