A coupled climate-ice sheet model is used in an attempt to simulate the northern hemispheric ice-sheet volume variations implied by deep-sea core records over the last several 105 years. The climate model consists of a global one-level diffusive energy-balance equation with seasonal variation and land-ocean contrast, using a severely truncated series of associated Legendre functions. The ice-sheet model uses a standard vertically averaged flow law and predicts ice thickness in a north-south cross section: bedrock deformation under the ice is included as linear flow confined to a thin asthenospheric channel. The climate model is tuned to the modern climate and tested against results from atmospheric general circulation models for 18,000 and 10,000 yr B.P. It is suggested that the lack of sea-ice variations causes the biggest departures from the GCM results. The sensitivities to the orbital perturbations of the coupled model and of the ice-sheet model alone with a prescribed snow-budget distribution are examined. Sone qualitative differences are found in the latitudinal forms of these sensitivities. Despite these differences, however, the two types of models produce similar ice-age simulations over the last 700 thousand years in response to the orbital perturbations. In particular, fairly realistic simulations with strong 100 Kyr cycles similar to those obtained by adding an ice-calving mechanism to the ice-sheet model in Pollard (1982a) are again produced in the coupled model.