We present one of the first estimates of the global distribution of CO<sub>2</sub> surface fluxes using total column CO<sub>2</sub> measurements retrieved by the SRON-KIT RemoTeC algorithm from the Greenhouse gases Observing SATellite (GOSAT). We derive optimized fluxes from June 2009 to December 2010. We estimate fluxes from surface CO<sub>2</sub> measurements to use as baselines for comparing GOSAT data-derived fluxes. Assimilating only GOSAT data, we can reproduce the observed CO<sub>2</sub> time series at surface and TCCON sites in the tropics and the northern extra-tropics. In contrast, in the southern extra-tropics GOSAT X<sub>CO<sub>2</sub></sub> leads to enhanced seasonal cycle amplitudes compared to independent measurements, and we identify it as the result of a land–sea bias in our GOSAT X<sub>CO<sub>2</sub></sub> retrievals. A bias correction in the form of a global offset between GOSAT land and sea pixels in a joint inversion of satellite and surface measurements of CO<sub>2</sub> yields plausible global flux estimates which are more tightly constrained than in an inversion using surface CO<sub>2</sub> data alone. We show that assimilating the bias-corrected GOSAT data on top of surface CO<sub>2</sub> data (a) reduces the estimated global land sink of CO<sub>2</sub>, and (b) shifts the terrestrial net uptake of carbon from the tropics to the extra-tropics. It is concluded that while GOSAT total column CO<sub>2</sub> provide useful constraints for source–sink inversions, small spatiotemporal biases – beyond what can be detected using current validation techniques – have serious consequences for optimized fluxes, even aggregated over continental scales.