The distributions of/i180, salinity, temperature, and nutrients have been used to quantify water sources to the Mackenzie shelf in the Beaufort Sea. Comparison of water mass analyses with satellite imagery confirms that the meteoric (runoff) water is associated with the Mackenzie plume. The seasonally variable surface layer for the shelf is viewed as cycling between a "reverse estuary" in winter, when the polar mixed layer (PML) is formed, and a positive estuary in summer when the shelf waters respond to freshwater inputs (runoff and ice melt). We infer a standing stock of 3.7 m fresh water at the end of summer 1986, of which 30% owes its origin to the melting of sea ice; our data coupled with river flow imply a freshwater flushing time for the Mackenzie shelf at about 150 days. To re-form the PML during winter requires the removal of this seasonal fresh water through the combined processes of flushing and ice formation: once this fresh water has been removed, continued ice growth can produce "new" brine which would be observed as a deeper and saltier PML from the previous year. A simple geochemical model shows that autumn conditions (freshwater accumulation) and the rate of flushing are important controls on the potential of the shelf to produce "new" brine and that winter runoff, were it to distribute evenly across the shelf, is sufficient to inhibit brine production. INTRODUCTION Arctic Ocean studies have recently focused on freshwater budgets [Aagaard and Coachman, 1975; Hanzlick and Aagaard, 1980; Anderson et al., 1983] and particularly on the two main sources, runoff (including Pacific Ocean sources) and ice melt [e.g., Tan and Strain, 1980; Ostlund and Hut, 1984]. The reason for this interest is simple: fresh water plays a key role in maintaining the Arctic halocline and, by inference, controls ice cover and deep water formation, both of which have climatic implications [Carmack, 1989]. The Arctic Ocean can be considered as a large, complex estuary where seasonal runoff impinges on wide shelves synchronously around the basin edge; in fact, the combined inflow (about 3500 km3yr -• [Treshnikov, 1985]) is surpassed in magnitude only by the Amazon. Additional runoff (estimated at 1670 km3yr -1 when referenced to salinity of 34.8 practical salinity units (psu) [Aagaard and Carmack, 1989]) enters indirectly through Bering Strait. Unique to cold region estuaries is a melt-freeze cycle which can cause positive or negative estuarine forcing [Carmack et al., this issue]. The logical place to look for reverse estuarine behavior (surface production of brine) is in the shear zone between landfast ice and the moving pack and in polynyas where water is kept open for long periods in winter. There seasonal processes have been implicated in producing brine drainage and hence maintaining the halocline [Melling and Lewis, 1982; Aagaard et al., 1981]. We expect therefore that the important estuarine processes (both negative and positive) occur over the shelves, are strongly seasonal, and are unevenly distributed within the Arc...