Atmospheric transport models are a source of uncertainty in the diagnostics of the CO2 sources and sinks. We propose here a protocol to compare two transport models: a 2-dimensional (2D) and a 3-dimensional (3D) model, based on 3 different experiments that reveal the ability of each model to account for the different components of the atmospheric carbon cycle. The 2D model we use is the one described by Tans et al. and the 3D model is the TM2 model, developed by Heimann et al. First, we conduct the same fossil fuel experiment in both models and show that the 2D model has a stronger interhemispheric mixing than the 3D model (∼ 25%), even though the 2D model presents a weaker intra-hemispheric mixing above source regions (experiment A). The influence of year-to-year variability of transport on the latitudinal profile in fossil-fuel CO2 appears to be weak for the 1990s. We then use a set of “all but fossil fuel” fluxes, originally inferred from the 2D model, as an input to the 3D model (experiment B). Even if the main discrepancy on the resulting latitudinal CO2 concentrations occurs between the 2D and 3D models in the tropics and at the mid-northern latitudes, the differences implied by three longitudinal distributions tested in the 3D model are important and can be explained by a few global transport mechanisms. Finally, we quantify the differences in latitudinal CO2 concentrations observed in experiment B in terms of net carbon fluxes at the surface. To do so, an inverse calculation of the CO2 fluxes in latitude and time is performed with the 3D model, using as an input a smoothed latitudinal profile of atmospheric measurements for the period 1990–1993 (experiment C=A + B). We find with the 3D model that, averaged on the period 1990–1993, the equatorial release is reduced by 40 Tmol yr−1 (roughly 25% of the original source) compared with the initial 2D budget and is shifted southward by roughly 10°. The mid northern latitude sink is also reduced by 80 Tmol yr−1 (roughly 25% of the original sink). In summary, this study shows that the changes in the carbon budget required when moving from the 2D model to this 3D model are important, but they are not radical changes.