A 5-year LDAS retrospective forcing data set (September 1996 - September 2001) was produced by the NASA/GSFC group for the North American Land Data Assimilation System (N-LDAS) project, and is being used to test the N-LDAS land surface models. The forcing data set covers the LDAS domain with a 1/8° resolution. It was produced based on near-surface analysis/assimilation fields from the NCEP Eta Data Assimilation System (EDAS), actual rain gauge and radar observations of precipitation, and satellite retrievals of downward solar radiation. The quality of the forcing is a critical factor that affects the evaluation of the participating land surface models. Here we validate this data set against independent observations from the Oklahoma Mesonet, ARM/CART facilities, and SURFRAD facilities. Meteorological sites at the Oklahoma Mesonet and ARM/CART Surface Meteorological Observation System (SMOS) have observations of 2 m air temperature, surface pressure, 10 m wind, and relative humidity. The incident solar radiation is also observed at these Oklahoma Mesonet sites, at ARM/CART Solar Infrared Radiation Stations (SIRS), and at 6 SURFRAD sites. SIRS also measures downward longwave radiation. Comparison between the N-LDAS forcing and the local observations shows a very good agreement for most of the variables. The importance of the sizes of the differences can be evaluated by examining the sensitivity of the land surface models to these differences. In addition to the original N-LDAS retrospective simulations, which used the standard N-LDAS forcing, we conducted additional simulations using the locally observed forcing. The results show that pairs of simulations with the different N-LDAS models give very similar soil moisture and surface energy fluxes for the different sets of forcings when suitably averaged, say over the state of Oklahoma, or over time. While there are certain differences on an hourly or daily time scale, especially because of differences in precipitation forcing, these differences do not produce long-term drifts in the simulations. This experiment shows that the current N-LDAS retrospective forcing is very close to the actual local forcing, and cannot explain any differences between observed and simulated soil moisture or fluxes.