We present a comparison of Arctic Ocean hydrographic data sets from the 1990s, with a focus on changes in the upper few hundred meters of the Eurasian Basin. The most recent observations discussed here were collected during the spring 1995 Scientific Ice Expedition (SCICEX'95), the second in a series of scientific cruises to the Arctic Ocean aboard U.S. Navy nuclear submarines. Although the 1990s have seen an abundance of synoptic cruises to the Arctic, this was the only one to take place in winter/spring conditions. Other data considered here were collected during the first SCICEX cruise in summer 1993 (SCICEX'93) and during an icebreaker cruise to the Eurasian Basin in summer 1991 (Oden'91). A new Russian-American winter climatology is also used as a reference. These comparisons reveal that the Eurasian Basin cold halocline layer has retreated during the 1990s to cover significantly less area than in previous years. Specifically, we find a retreat from the Amundsen Basin back into the Makarov Basin; the latter is the only region with a true cold halocline layer during SCICEX'95. Changes are also seen in other halocline types and in the Atlantic Water layer heat content and depth. Since the cold halocline layer insulates the surface layer (and thus the overlying sea ice) from the heat contained in the Atlantic Water layer, this should have profound effects on the surface energy and mass balance of sea ice in this region. Using a simple mixing model, we calculate maximum ice-ocean heat fluxes of 1--3 W m-2 in the Eurasian Basin, where during SCICEX'95 the surface layer lay in direct contact with the underlying Atlantic Water layer. The overall cause of water mass changes in the 1990s might have been a shift in the atmospheric wind forcing and resulting sea ice motion during the late 1980s, which we speculate influenced the location where fresh shelf waters flow into the deeper basins of the Arctic Ocean. Finally, we discuss two different mechanisms that have been proposed for cold halocline water formation, and we propose a compromise that best fits these data. ¿ 1998 American Geophysical Union