A two-source model (TSM) for surface energy balance, considering explicitly soil and vegetation components, was tested under water stress conditions. The TSM evaluated estimates the sensible heat flux (H) using the surface-air thermal gradient and the latent heat flux (LE) as a residual from the surface energy balance equation. The analysis was performed in a semiarid Mediterranean tussock grassland in southeast Spain, where H is the dominant flux and LE rates are low, challenging conditions under which the TSM has not been validated before. We evaluated two different resistance schemes: series and parallel; as well as the iterative algorithm included in the TSM to disaggregate the soil-surface composite temperature into its separate components. Continuous field measurements of composite soil-vegetation surface temperature (T) and bare soil temperature (T) from thermal infrared sensors were used for model testing along with canopy temperature estimates (T'), derived from T and T.Comparisons with Eddy covariance and field data showed that the TSM produced reliable estimates of net radiation (Rn) and H fluxes, with errors of ~30% and ~10%, respectively, but not for LE, with errors ~90%. Despite of lower errors (~10%) in estimating H using parallel resistance, the series scheme was more robust showing slightly higher correlations (r=0.78-0.80 vs. r=0.75-0.77) and allowing a better disaggregation of soil and canopy fluxes. Differences between model runs using the iterative algorithm to disaggregate T and the simplified version that uses separate inputs of T and T' were minor. This demonstrates the robustness of the iterative procedure to disaggregate a composite soil-vegetation temperature into separate soil and vegetation components in semiarid environments with good prospects for image applications.