Numerical land surface models (LSMs) are abundant and in many cases highly sophisticated, yet their output has not converged towards a consensus depiction of reality. Addressing this matter is complicated by the huge number of possible combinations of input land characteristics, forcings, and physics packages available. The Global Land Data Assimilation System (GLDAS) and its sister project the Land Information System (LIS) have made it straightforward to test a variety of configurations with multiple LSMs. In order to compare the impacts of the choice of LSM, land cover, soil, and elevation information, and precipitation and downward radiation forcing datasets on simulated evapotranspiration, sensible heat flux, and top layer soil moisture, a set of experiments was designed which made use of high quality, physically coherent, 1-year datasets from four reference sites of the Coordinated Enhanced Observing Period (CEOP) initiative. As in previous studies, it was shown that the LSM itself is generally the most important factor governing output. Beyond that, evapotranspiration seems to be most sensitive to precipitation, land cover, and radiation (in that order); sensible heat flux is most sensitive to radiation, precipitation, and land cover; and soil moisture is most sensitive to precipitation, soil, and land cover. Various seasonal and model specific dependencies and other caveats are discussed. Output fields were also compared with observations in order to test whether the LSMs are capable of simulating an observed reality given a plausible set of inputs. In general, that potential was fair for evapotranspiration, good for sensible heat flux but problematic given its strong sensitivity to the inputs, and poor for soil moisture. The results emphasize that improving the LSMs themselves, and notjust the inputs, will be essential if we hope to model land surface water and energy processes accurately.