This paper proposed a new pixel decomposition model of Temperature Unmixing with Spectral (TUS). Landsat TM data acquired in Beijing were used for the study. Firstly, land surface fraction was obtained based on the Linear Spectral Mixing Model.Secondly and LST of typical endmember was selected through Temperature Vegetation Index. Finally, pixel decomposition of LST can be achieved integrated emissivity with different surface components. Our results indicated that TUS can effectively improve the spatial resolution of land surface temperature, reflecting the spatial differences of surface components, with MAE and RMSE 1.25K and 2.27K respectively. Therefore we conclude that TUS model is applicable for decomposition of LST images for high spatial resolution in the complex surface coverage area.