A two-dimensional model which links the atmosphere, the mixed layer of the ocean, the sea ice, the continents, the ice sheets and their underlying bedrock has been used to test the Milankovitch theory over the last two glacial-interglacial cycles. A series of sensitivity analyses have allowed us to understand better the internal mechanisms which drive the simulated climate system and in particular the feedbacks related to surface albedo and water vapour. It was found that orbital variations alone can induce, in such a system, feedbacks sufficient to generate the low frequency p art of the climatic variations over the last 122 ka. These simulated variations at the astronomical timescale are broadly in agreement with reconstructions of ice-sheet volume and of sea level independently obtained from geological data. Imperfections in the stimulated climate were the insufficient southward extent of the ice sheets and the too small hemispheric cooling at the last glacial maximum . These deficiencies were partly remedied in a further experiment by using the time-dependent atmospheric CO2 concentration given by the Vostok ice core in addition to the astronomical forcing. In this transient simulation, 70% of the Northern Hemisphere ice volume is related to the astronomical forcing and the related changes in the albedo, the rem aining 30% being due to the CO
2
changes. Analysis of the processes involved shows that variations of ablation are more important for the ice-sheet response than are variations of snow precipitation. A key mechanism in the deglaciation after the last glacial maxim um appears to be the ‘ageing’ of snow which significantly decreases its albedo. The other factors which play an important role are ice-sheet altitude, insolation, taiga cover, ice-albedo feedback, ice-sheet configuration (‘continentality’ and ‘desert’ effect), isostatic rebound, CO
2
changes and tem perature-water vapour feedback. Numerical experiments have also been carried out with a one-dimensional radiative-convective model in order to quantify the influence of the CO
2
changes and of the water vapour feedback on the climate evolution of the Northern Hemisphere over the last 122 ka. Results of these experiments indicate that 67% of the simulated cooling at the last glacial maximum can be attributed to the astronomical forcing and the subsequent surface albedo increase, the remaining 33% being associated with the reduced CO
2
concentration. Moreover, the water vapour feedback explains 40% of the simulated cooling in all the experiments done. The transient response of the clim ate system to both the astronomical and CO
2
forcing was also simulated by the LLN (Louvain-la-Neuve) 2.5-dimensional model over the two last glacial-interglacial cycles. It is particularly significant that spectral analysis of the simulated Northern Hemisphere global ice volume variations reproduces correctly the relative intensity of the peaks at the orbital frequencies. Except for variations with timescales shorter than 5 ka, the simulated long-term variations of total ice volume are comparable to that reconstructed from deep sea cores. For example, the model simulates glacial maxima of similar amplitudes at 134 ka BP and 15 ka BP, followed by abrupt deglaciations. The complete deglaciation of the three main Northern Hemisphere ice sheets, which is simulated around 122 ka BP, is in partial disagreement with reconstructions indicating that the Greenland ice sheet survived during the Eemian interglacial. The continental ice volume variations during the last 122 ka of the 200 ka simulation are, however, not significantly affected by this shortcoming.
Climate response to freshwater perturbations in Northern or Southern Hemispheres at the last glacial inception, the last glacial maximum and the present-day