Dense water formation from coastal polynyas on the Chukchi Shelf is examined using a primitive-equation ocean model forced by surface buoyancy fluxes from a time-dependent polynya model for the winters of the 1978–1998 period. The model is forced by either meteorological observations or National Center for Environmental Prediction (NCEP) reanalysis data. During the 21-year period the surface forcing and the dense water production vary by a factor of 2. Using meteorological observations, most winters produce substantial amounts of water with a density anomaly of at least 1.2 kg m−3 (salinity increase of 1.5 psu), while using NCEP forcing, density anomalies typically reach 0.8 kg m−3 (1.0 psu) but are rarely as large as 1.2 kg m−3. With meteorological forcing, the production of dense water is fairly uniform throughout the entire period, whereas with NCEP forcing, nearly all of the water with density anomaly greater than 1.2 kg m−3 is formed during the early part of the modeled period (1978–1984), with the density anomaly rarely reaching 1 kg m−3 in the later part (1987 and onwards). Interannual variability is high with large differences between successive winters in both ice production and dense water production. Most of the observed variability can be explained by varying wind fields, with offshore winds creating polynyas between 30 and 75% of the time during November to April. Using a climatologically based mean initial salinity of 31.6 psu, we show that maximum salinities produced rarely exceed 33.5 psu. Furthermore, on the basis of moored observations in Bering Strait, we conclude that the interannual variability of the initial salinity is of the same magnitude as the interannual variability in dense water formation, and thus both are equally important in determining whether or not winter water is dense enough to contribute to the cold halocline layer of the Arctic Ocean. Winters with high fractions of offshore winds can produce anomalies up to 1.8 kg m−3 (2.2 psu). This, together with the varying initial salinity (density) fields, can achieve waters with maximum salinities up to 35.4 psu. Finally, we find that the derived ice volumes and dense water productions are highly sensitive to the forcing (meteorological vs. NCEP), and comparison with in situ observations is highly recommended.