The terrestrial biosphere is one of several key components of the global carbon cycle. Because the mechanisms by which climate determines terrestrial biosphere carbon fluxes are not well understood, significant uncertainties concerning model results exist even for the current state of the system, with important consequences for our ability to predict changes under future climate change scenarios. We assess how far this uncertainty can be reduced by constraining a global mechanistic model of vegetation activity, either with global satellite-derived vegetation index data or with measurements of the seasonal CO2 cycle in the atmosphere. We first show how constraining the model with satellite data from the National Oceanic and Atmospheric Administration advanced very high resolution radiometer reduces the sensitivity to estimated uncertainties in model parameters, and thus the estimated error range of net primary productivity. Regionally, the satellite data deliver the largest constraint for vegetation activity in boreal and arctic as well as in tropical water-limited environments. In a second analysis through an atmospheric tracer transport model, we check the consistency of those results with the measured seasonal cycle of CO2 at various remote monitoring sites. While before including the satellite data into model calculations, some simulations within the error range lead to a CO2 seasonal cycle outside the observations, there is a good agreement with the additional constraint. The conclusion is that the constraint delivered by the satellite data is at least as significant as that delivered by atmospheric CO2 measurements. We also show that the CO2 data mainly reflect the activity of northern vegetation, in particular conifers and C3 grasses. This suggests that satellite measurements provide the most useful global data currently available for checking and improving terrestrial vegetation models and that consistency with CO2 measurements is a necessary but not a sufficient requirement for their realism.