Data derived at the National Centers for Environmental Prediction via four-dimensional data assimilation using the Eta Model were evaluated against surface observations from two observational arrays, one located in the semihumid, continental climate of Oklahoma and Kansas and the second in the semiarid climate of southern Arizona. Comparison was made for the period of the Global Energy Water-cycle Experiment Continental-scale International Project’s ‘‘GIST’’ dataset in 1994 and their ‘‘ESOP-95’’ dataset in 1995, and for the months of March and May in 1996. Coding errors in the Eta Model’s postprocessor used to diagnose near-surface temperature and humidity are shown to have compromised the GIST and ESOP-95 near-surface data. A procedure was devised to correct the GIST and ESOP-95 near-surface fields by mimicking the corrected code used in the Eta Model since January 1996. Comparison with observations revealed that modeled surface solar radiation is significantly overestimated except in clear-sky conditions. This discrepancy in cloudy-sky solar radiation was altered little by the substantial January 1996 revisions to Eta Model physics, but the revisions are shown to have greatly improved the model’s ability to capture daily and seasonal variations in near-surface air temperature, specific humidity, and wind speed. The poorly modeled surface radiation complicates evaluation of modeled surface energy fluxes, but comparison with observations suggests that the modeled daytime Bowen ratio may be systematically high. This study clearly demonstrates the strong sensitivity of model-calculated, near-surface variables to the physics used to describe surface interactions in the data assimilation model. To mitigate against this and to aid intercomparisons between other data, it is recommended that model-derived data always include sufficient information to allow potential users to recalculate the extrapolation to the surface using a user-defined model of surface‐atmosphere exchanges.