Among the issues raised by the globally ice-covered Earth, or a so-called 'hard' snowball-Earth scenario, one of the most important is to establish the CO 2 threshold required for the deglaciation. This problem has been addressed using the Energy-Balance Model (or EBM), which showed that for Neoproterozoic insolation, 0.29 bar of CO 2 would be needed to trigger deglaciation. New results, obtained with the Atmospheric General Circulation Model (AGCM) FOAM (for Fast Oceanic Atmospheric Model), have demonstrated that, even with an atmospheric content of 0.2 bar, the equatorial temperature remained far below the level required for the deglaciation. Those results show that the cause of deglaciation is unresolved and the discussion about a plausible escape scenario remains open. For this reason, to test and to determine the sensitivity and efficiency of the greenhouse effect during a 'hard' snowball-Earth, we compare the FOAM results with those of LMDz (AGCM of the 'Laboratoire de météorologie dynamique'). The preliminary results show that LMDz is much more sensitive to a CO 2 increase than FOAM. This article shows that among processes that could explain this difference, the key factor is the cloud parameterization and its interaction with the convective scheme. These simulations suggest that the CO 2 threshold is dependent on the GCM parameterization used, and could be lower than the one suggested by FOAM. Moreover, to investigate other plausible mechanisms able to melt the equatorial ice, we have tested the CH 4 impact with a simple 0D model, INCA-ZD. Results show that the balance between the residence times of CH 4 in a 'hard' snowball-Earth scenario is largely overcome by the extinction of the organic source, which means that CO 2 remains the only greenhouse gas warming the snowball Earth.