Difficulties of emissivity determination and atmospheric correction degrade the estimation accuracy of land surface temperature (LST). That is, since the emissivity determination of land surface material and the correction of atmospheric effect are not perfect, it is very difficult to estimate the precise LST from a thermal infrared image such as Landsat TM and ETM+, ASTER, etc. In this study, we propose an efficient method to estimate land surface temperature difference (LSTD) rather than LST from Landsat thermal band images. This method is based on the assumptions that 1) atmospheric effects are same over a image and 2) the emissivity of vegetation region is 0.99. To validate the performance of the proposed method, error sensitive analysis according to error variations of reference land surface temperature and the water vapor is performed. The results show that the estimated LSTD have respectively the errors of ${pm}0.06K$, ${pm}0.15K$ and ${pm}0.30K$ when the water vapor error of ${pm}0.302g/cm^2$ and the radiance differences of 0.2, 0.5 and $1.0Wm^{-2}sr^{-1}{mu}m$ are considered. And also the errors of the LSTD estimation are respectively ${pm}0.037K$, ${pm}0.089K$, ${pm}0.168K$ in the reference land surface temperature error of ${pm}2.41K$. Therefore, the proposed method enables to estimate the LSTD with the accuracy of less than 0.5K.