Subsea sonars moored in the Beaufort Sea acquired a spatial section of draft across 941 km of sea ice during the winter of 1991--1992. These observations document the development of seasonal sea ice from an open sea surface in October to an average accumulation exceeding 2.5 m draft by early April. Initially, level ice occupied 85% of the profile, but continued ridging reduced this fraction to 50% in late winter. Three modes of level ice were tracked: that of seasonal ice whose growth was initiated at freeze-up, and two others created during openings of the ice field in January and in March. The growth of these modes can be closely matched by calculations based on a slab model of ice formation forced using surface meteorological data. Initially, ridge keels were small and widely spaced. By midwinter an exponential dependence of keel frequency on draft was obvious (e-folding scale, 2.16 m), and the frequency and mean draft of keels had stabilized. The maximum keel draft was a modest 17.4 m. The low incidence of deep keels at all times is a statistically significant departure from the exponential dependence valid at lesser drafts. The truncation point of the exponential relation is related to the draft of the thinnest level ice present. In late winter the formation of deeper keels from thick first-year ice was apparently precluded by the presence of younger, thinner ice, which limited the force available for ridge building. Through calculation of the seaward transport of ice over the sonars, the total production of ice in the coastal flaw lead during 1991--1992 was determined to be about 60% of earlier indirect estimates. In general, the observations revealed winter ice conditions significantly less severe than those found on the periphery of the polar pack only a few tens of kilometers to the north, but more so than conditions in a marginal first-year ice zone at the same latitude in Davis Strait. ¿ American Geophysical Union 1996