By using an observation system simulation experiment, the simulated soil moisture, soil temperature and surface heat fluxes are firstly compared for a period of soil drying process, respectively, from an offline Land Surface Model (LSM) and a Single Column Model (SCM) with coupled atmospheric boundary layer and LSM, and secondly, the ability of estimated land surface states are evaluated from the two models with the  aid of data assimilation technique. The results show the differences of all the simulated variables are small except long-wave radiation and sensible heat fluxes being relatively large. For only the near-surface soil moisture observation assimilation, the estimated soil moisture and surface sensible heat flux with LSM are generally better than those with SCM, on the contrary, the estimated soil temperature with SCM is better, and the difference of estimated latent heat fluxes is small. However, the assimilation of both near-surface soil temperature and soil moisture observations at the same time will make the difference of heat flux estimates become large from the two models by the end of data assimilation cycle. Finally, the possible reasons for the different performances of the two models are investigated and analyzed by using the sensitivity tests. The above results imply that it should be careful when we use one-model output as another-model input, and avoid the dynamic inconsistency between the two-model states.