The sensitivity of a global thermodynamic-dynamic sea ice model coupled to a one-dimensional upper ocean model to degradations of the model physics is investigated. The thermodynamic component of the sea ice model takes into consideration the presence of snow on top of sea ice, the storage of sensible and latent heat inside the snow-ice system, the influence of the subgrid-scale snow and ice thickness distributions on sea ice thermodynamics, the transformation of snow into snow ice when snow depth increases to the point where the snow-ice interface sinks below the waterline, and the existence of leads and polynyas (areas of open water) within the ice cover. Ice dynamics is treated basically as by Hibler <1979>. Regarding the upper ocean model, it is made up of an integral mixed layer model and of a diffusive model of the pycnocline. Advection of heat and salt by oceanic currents is implicitly accounted for by restoring the temperatures and salinities of the water column to annual mean data. It is very important to note that a single set of parameter values is employed to simultaneously simulate the Arctic and Antarctic ice regimes. A control run carried out with the model demonstrates that it does reasonably well in simulating the seasonal waxing and waning of both ice packs. The sensitivity study focuses on physical processes pertaining to (1) the vertical growth and decay of sea ice (thermal inertia of snow and ice, heat conduction, and snow cover), (2) the lateral growth and decay of sea ice (leads and polynyas), and (3) the sea ice dynamics (ice motion and shear strength). A total of nine sensitivity experiments have been performed. Each experiment consisted of removing a particular parameterization from the control run computer code. It appears that the thermal inertia of the snow-ice system is negligible in the Antarctic but not in the Arctic, where the total heat content of sea ice is chiefly dictated by internal storage of latent heat in brine pockets, sensible heat storage being of very minor consequence. It is also found that the inclusion of a prognostic snow layer and of a scheme of snow ice formation is important for sea ice modeling in the southern hemisphere. Furthermore, our results suggest that the thermodynamic effect of the subgrid-scale snow and ice thickness distributions, the existence of open water areas within the ice cover, and the ice motion play a crucial role in determining the seasonal behavior of both ice packs. The ice shear strength seems to be of lesser importance, although it has a nonnegligible effect in both hemispheres. We can therefore conclude that all these processes should be represented in global climate models.¿ 1997 American Geophysical Union