To improve the accuracy of atmospheric boundary layer simulation, we propose and have validated an inverse technique for assimilating soil moisture using surface observations. Soil moisture profoundly affects mesoscale model simulations. Despite its importance, it is one of the most uncertain variables to be initialized in a numerical model. Further, the errors in the estimated initial soil moisture are persistent and are not reduced within a typical case study/episode. We followed a thermodynamical approach to perform a continuous data assimilation of surface-layer temperature and water vapor mixing ratio to develop dynamically consistent surface-atmosphere coupling. First, we used the analyzed surface data to directly assimilate surface-layer temperature and water vapor mixing ratio. We then used the difference between the observations and model predictions to calculate adjustments to the surface fluxes of sensible and latent heat. These adjustments were used to calculate a new estimate of the ground temperature, thereby affecting the predicted surface fluxes in the subsequent time step. We also used the adjusted latent heat fluxes to calculate changes in soil moisture using an inverse method. This indirect data assimilation of ground temperature and soil moisture was applied simultaneously with the direct assimilation of surface data in the model's surface layer, thereby maintaining greater consistency between the surface layer mass-field variables and the ground temperature and moisture. This technique was tested for two different land surface models: a diagnostic scheme based on land use and prespecified soil moisture availability, Ma, following Carlson and Boland as used in the Mesoscale Model Version 5 (MM5); and a prognostic scheme suggested by Noilhan and Planton that is being used in many mesoscale land surface models. In the first case, we showed that direct and indirect assimilation of surface observations can be efficiently used to modulate the Ma term and thus develop more realistic soil moisture feedback in the Carlson and Boland scheme. For the second (Noilhan and Planton) scheme, this assimilation approach was used to estimate adjustments for equitable evapotranspiration which were then translated into corresponding soil moisture adjustments in the soil moisture prognostic equations. Our technique was validated using a 1-D model along with special field observations over different landscapes with a variety of vegetation and soil moisture stress levels, and was found to be an efficient approach that can be easily adopted in various mesoscale models.