This paper investigates the impact of seasonal covariances between transport and concentrations on inverse estimates of CO2 sources with an atmospheric transport model. We use two atmospheric transport models differing only in their parameterizations of subgrid vertical diffusion. The models have similar transport characteristics for nonseasonal sources but different covariances between seasonal sources and transport. Four mass balance inversions have been performed with different model combinations to estimate 1981--1996 CO2 sources. The cases indicate typical and upper bound impacts due to seasonal covariance and observing network size. Typical differences are small on semihemispheric scales for a network of approximately 20--25 oceanic and coastal sites. Upper bound differences are large, particularly in the partitioning of sources between land and ocean. This behavior is model dependent. The interannual variability of the four inversions is similar. This suggests that estimates of interannual variability of sources are more robust to uncertainties in transport than long-term mean estimates. ¿ 1999 American Geophysical Union