Single point simulation experiments have been conducted using NOAH (The Community Noah Land Surface Model), SHAW (simultaneous Heat and Water) and CLM (Community Land Model) respectively. Results are compared with observed data from Semi-Arid Climate Observatory and Laboratory (SACOL) to study the land surface process in the Loess Plateau. The goal of this paper is to evaluate the simulation capability of different models in semi-arid areas by comparing model simulations with observations. The results show large differences in the performance of the three models. For soil temperature at the top 20 cm, the simulation capability of CLM 4.0 is the best, while the performance of SHAW is the best for simulation of temperature in deep soil layers. For soil water content, the simulated results of SHAW are the closest to observed values, while large biases are found in simulations of NOAH and CLM. All the three models can well simulate upward shortwave radiation, and the performance of SHAW is the best. The performance of CLM for upward longwave radiation simulation is the best. All the three models can simulate variations of sensible and latent heat fluxes, and the performance of CLM for sensible heat flux simulation is better than that of NOAH and SHAW. As to latent heat flux, CLM performance is the best under wet condition after precipitation, whereas there are large biases in CLM simulation under dry condition without precipitation. The performance of NOAH for latent heat flux simulation is the best in the winter. Relatively, CLM can better represent the interaction between land and atmosphere in semi-arid areas, but its performance for soil water content and latent heat flux simulation under dry condition is not ideal. Further research and refinement for the hydrological process between land and atmosphere described in CLM is needed in semi-arid areas.