A fully prognostic coupled ice-ocean model is used to examine the importance of interannually variable ocean forcing, on large-scale monthly simulations of the Arctic ice cover. The model is forced by a prescribed interannually variable atmosphere. Simulations are conducted for a 120-month period and validated with observed ice concentration data. The model produces very reasonable simulations of the ice edge position, particularly in the Barents and Greenland seas. The use of interannual ocean forcing produces major improvements to the simulation of the ice concentration for both the annual cycle and interannual variations, as compared with simulations in which the ocean forcing is a prescribed mean annual cycle. Vertical ocean heat flux appears to be the dominant mechanism controlling localized ice area anomalies and the overall ice concentration. Consistent errors in simulated ice concentration and thickness remain, including a slightly exaggerated melt-freeze cycle and insufficient ice thickness. The lack of a nonprognostic mixed layer and the coarse vertical resolution are apparently inadequate to represent the vertical mixing, stratification, and diffusion processes properly.