Remotely sensed thermal inertia method has been recognized as a promising approach for land surface soil moisture retrieval from the early 1970's. In order to estimate the land surface soil moisture in arid regions, a real thermal inertia (RTI) model was formulated based on the heat conduction equation and an approximated energy budget equation at the land surface using the land surface temperature and reflectance measured by Moderate Resolution Imaging Spectroradiometer (MODIS). The soil thermal inertia of Heihe River Basin (HRB) was retrieved based on the RTI model. Furthermore, using a thermal inertia-soil moisture model along with auxiliary data such as soil texture and bulk density, land surface soil moisture was estimated. The results were verified experimentally using the observations made at three automatic weather stations (AWS). The coefficient of the correlation between the retrieved values of soil thermal inertia and measured ones was with above R=0.6 and the root mean square error of soil moisture was 0.072 m3m-3. The soil moisture in the HRB exhibits a seasonal variation with higher values in summer and autumn and lower values in winter and spring, and also exhibits considerable spatial variation with higher values in the upstream district and lower values in the downstream district.