It is well documented that soil moisture can be retrieved from passive microwave (PMW) observations. A basic assumption of most PMW-based SM retrieval algorithms is that vegetation temperatures ( Tv ) and soil temperatures ( Ts ) are equal (i.e., Tv=Ts ), which however is not well satisfied in some cases, especially during daytime. In this study, we proposed a soil-vegetation temperature decomposition (SVTD) approach to avoid such an assumption, which can improve the accuracy of SM retrievals from the Advanced Microwave Scanning Radiometer 2 (AMSR2) data. First, the SVTD was used to decompose the vegetation and soil temperatures of the soil-vegetation mixed pixels in the Tibetan Plateau (TP). Subsequently, the decomposed temperature was integrated into the SM retrieval algorithm to correct the effects of soil and vegetation temperatures, and SM is then retrieved following the same strategy adopted in the land parameter retrieval model (LPRM). Finally, the algorithm was validated against densely instrumented SM networks (Maqu, Naqu, and Ngari) built in the TP and was also compared with the LPRM AMSR2 SM product. Results indicate that the proposed algorithm performs much better than the original LPRM in soil-vegetation mixed areas. The proposed SVTD method is promising for SM retrieval from PMW satellites, especially in the daytime when the difference between soil and vegetation temperatures is relatively large.