We present results from numerical experiments made with a GCM, the NCAR CCM1, that were designed to estimate the annual balance between snow-fall accumulation and ablation for geographically important land regions for a variety of conditions. We also attempt to assess the reliability of these results by investigating model sensitivity to changes in prescribed physical parameters. Experiments were run with an initial imposition of 1 m of (midwinter) snowcover over all northern hemisphere land points. Over Alaska, western Canada, Siberia, and the Tibetan Plateau the model tended to retain this snow cover through the summer and in some cases increase its depth as well. We define these regions as “glaciation sensitive” and note some correspondence between them and source regions for the Pleistocene ice sheets. An experiment with greatly reduced CO2 (100 ppm) showed a tendency towards “spontaneous” glaciation, i.e., the model remained snow-covered throughout the summer over the same geographic regions noted above. With 200 ppm CO2 (roughly equal to values at the last glacial maximum), snow cover over these regions did not quite survive the summer on a consistent basis. Combining 200 ppm CO2 and 1 m of initial northern hemisphere snow cover yielded glaciation-sensitive conditions, agreeing remarkably well with locations undergoing glaciation during the Pleistocene. To assess the reliability of these results, we have determined minimal model uncertainty by varying two of the empirical coefficients in the model within physically plausible ranges. In one case surface roughness of all ocean gridpoints was reduced by an order of magnitude, leading to local 10% reductions in precipitation (snowfall), a change hard to distinguish from inherent model variability. In the other case, the fraction of a land grid square assumed to be occupied by snow cover for albedo purposes was varied from one-half to unity. Large changes occurred in the degree of summer melting, and in some cases the sign of the net balance changed as fractional snow cover was changed. We conclude that the model may be able to reveal regions sensitive to glaciation, but that it cannot yield a reliable quantitative computation of the magnitude of the net snow accumulation that can be implicitly or explicitly integrated through time.