A coupled ice-ocean model of the Arctic is developed in order to study the effects of precipitation and river runoff on sea ice. A dynamic-thermodynamic sea ice model is coupled to an ocean general circulation model which includes a turbulent closure scheme for vertical mixing. The model is forced by interannually varying atmospheric temperature and pressure data from 1980-1989, and spatially varying mean monthly precipitation and river runoffs. Salinity and fresh water fluxes to the ocean from ice growth, snow melt, rain, and runoffs are computed, with no artificial constraints on the ocean salinity. The modeled ice thickness is similar to the observed pattern, with the thickest ice remaining against the Canadian Archipelago throughout the year. The modeled ice drift reproduces the Beaufort gyre and Transpolar drift exiting through Fram Strait. The stable arctic halocline produced by the vertical mixing scheme isolates the surface from the Atlantic layer and reduces the vertical fluxes of heat and salinity. A sensitivity experiment with zero precipitation results in rapidly decreasing ice thickness, in response to greater ocean heat flux from a weakening of the halocline, while an experiment with doubled precipitation results in a smaller increase in ice thickness. A zero-runoff experiment results in a slower decrease in ice thickness than the zero-precipitation case, due to the decadal time scale of the transport of runoff in the model. The results suggest that decadal trends in both arctic precipitation and river ru noffs, caused either by anthropogenic or natural climatic change, have the potential to exert broad-scale impacts on the arctic sea ice regime.