Atmospheric transport model errors are a major contributor to uncertainty in CO 2 inverse flux estimates. Our study compares CO 2 mole fraction observations from the North American Carbon Program Mid-Continental Intensive (MCI) field campaign and modeled mole fractions from two atmospheric transport models: the global Transport Model 5 from NOAA's CarbonTracker system and the mesoscale Weather Research and Forecasting model. Both models are coupled to identical CO 2 fluxes and lateral boundary conditions from CarbonTracker (CT2009 release). Statistical analyses were performed for two periods of 2007 using observed daily daytime average mole fractions of CO 2 to test the ability of these models to reproduce the observations and to infer possible causes of the discrepancies. TM5-CT2009 overestimates midsummer planetary boundary layer CO 2 for sites in the U.S. corn belt by 10 ppm. Weather Research and Forecasting (WRF)-CT2009 estimates diverge from the observations with similar magnitudes, but the signs of the differences vary from site to site. The modeled mole fractions are highly correlated with the observed seasonal cycle (r ≥ 0.7) but less correlated in the growing season, where weather-related changes in CO 2 dominate the observed variability. Spatial correlations in residuals from TM5-CT2009 are higher than WRF-CT2009 perhaps due to TM5's coarse horizontal resolution and shallow vertical mixing. Vertical mixing appears to have influenced CO 2 residuals from both models. TM5-CT2009 has relatively weak vertical mixing near the surface limiting the connection between local CO 2 surface fluxes and boundary layer. WRF-CT2009 has stronger vertical mixing that may increase the connections between local surface fluxes and the boundary layer.