We used continuous measurements of atmospheric CO2 at two stations to investigate potential errors in inversions of temporal averages of satellite clear-sky column retrievals. Compared to the complete data sets, the mid-day CO2 on clear days was systematically lower with a larger winter difference. Net ecosystem exchange (NEE) of CO2 was enhanced on clear vs. all days, the summer boundary layer was deeper, and the CO concentration was systematically lower. During winter these differences cannot account for the CO2 bias, which must be caused by advection. Summertime errors reflect a tradeoff between deeper mixing and enhanced NEE on clear days. If these sites represent mid-latitude forests and if the CO2 difference is confined to the bottom 15% column mass, then inversions of temporally-averaged satellite column data products will incur a -0.2 to -0.4 ppm bias. CO2 concentrations must therefore be assimilated at the place and time observed.