A global three-dimensional atmospheric model, the NCAR CCM2 general circulation model, has been adapted to study the hourly to yearly variability of CO2 in the atmosphere. Features of this CCM2-based model include high spatial resolution (2.8° × 2.8° latitude/ longitude), 18 vertical levels, a 15-min time step, and an explicit, nonlocal atmospheric boundary layer parameterization. The surface source/sink relationships used include exchange with the ocean, the terrestrial biosphere, biomass burning, and fossil fuel release of CO2. The timing and magnitude of the model seasonal cycle are compared to observational data for 28 sites. The seasonal cycle of atmospheric CO2 is generally well predicted by the model for most of the northern hemisphere, but estimates of the amplitude of the seasonal cycle in the southern hemisphere are overpredicted. To address this aspect more rigorously, we have used the monthly surface ocean pCO2 maps created by the Max-Planck-Hamburg ocean general circulation model to asses the ocean seasonality on the atmospheric surface CO2 seasonality. The globally averaged interhemispheric gradient in atmospheric CO2 concentrations, as computed with the chosen source/sink distributions, is a factor of two too high compared to data, and selected longitudinal bands may be up to 50% higher than the zonal mean. The high temporal resolution of this model allows the infrequent yet real extrema in atmospheric CO2 concentrations to be captured. The vertical attenuation of the seasonal cycle of atmospheric CO2 is well simulated by the boundary layer/free troposphere interaction in the model in the northern hemisphere. Conversely, an increasing amplitude of the seasonal cycle aloft is found in the midlatitude southern hemisphere indicating interhemispheric transport effects from north to south. We use two different models of the terrestrial biosphere to examine the influence on the computed seasonal cycle and find appreciable differences, especially in continental sites. A global three-dimensional chemical transport model is used to assess the production of CO2 from the oxidation of CO throughout the volume of the atmosphere. We discuss these CO + OH → CO2 + H results within the context of inverse model approaches to ascertaining the global and regional source/sink patterns of CO2. Deficiencies in the model output as compared to observational data are discussed within the context of guiding future research.