We estimate the annual net air-sea fluxes of oxygen for 13 regions on the basis of a steady state inverse modeling technique that is independent of air-sea gas exchange parameterizations. The inverted data consist of the observed oceanic oxygen concentration after a correction has been applied to account for biological cycling. We find that the tropical oceans (13¿S--13¿N) emit ~212 Tmol O2 yr-1, which is compensated by uptake of 148 Tmol yr-1 in the Northern Hemisphere (>13¿N) and by uptake of 65 Tmol yr-1 in the Southern Hemisphere (<13¿S). These results imply that the dominant feature of oxygen transport in the combined ocean-atmosphere system is the existence of a closed circulation cell in each hemisphere. These two cells consist of O2 uptake by the ocean in the middle and high latitudes of both hemispheres and transport in the ocean toward the tropics, where O2 is lost to the atmosphere and transported in the atmosphere back toward the poles. We find an asymmetry in the two cells involving O2 uptake in the temperate regions of the Northern Hemisphere versus loss of O2 in the temperate regions of the Southern Hemisphere. There is an additional asymmetry between the Atlantic basin, which has a net southward transport at all latitudes north of 36¿S, in agreement with independent transport estimates, versus the Indian and Pacific Oceans, which have a strong equatorward transport everywhere. We find that these inverse estimates are relatively insensitive to details in the inversion scheme but are sensitive to biases in the ocean general circulation model that provides the linkage between surface fluxes and ocean interior concentrations. Forward simulations of O2 in an atmospheric tracer transport model using our inversely estimated oxygen fluxes as a boundary condition agree reasonably well with observations of atmospheric potential oxygen (APO≈O2+CO2). Our results indicate that the north-south asymmetry in the strength of the two hemispheric cells coupled with a strong asymmetry in fossil fuel emissions can explain much of the observed interhemispheric gradient in APO. Therefore it might not be necessary to invoke the existence of a large southward interhemispheric transport of O2 in the ocean, such as proposed by Stephens et al. [1998]. However, we find that uncertainties in the modeled APO distribution stemming from seasonal atmospheric rectification effects and the limited APO data coverage prevent the currently available APO data from providing strong constraints on the magnitude of interhemispheric transport. ¿ 2001 American Geophysical Union