The riverine flux of U that enters the deep oceans is not well constrained since the net losses during estuarine mixing are difficult to quantify. Riverine-dissolved U normally has a higher U-234/(238) U activity ratio (U-234/U-238(ar)) than the uniform value that characterizes open ocean seawater and could be used as a tracer of riverine inputs if one could resolve subtle variations in seawater composition. Using new mass spectrometry techniques we achieve a long-term reproducibility +/- 0.3 parts per thousand on U-234/U-238(ar) which permits the tracing of riverine U in seawater samples from the Arctic - a partially restricted basin that is ideal for such a study. We find that surface waters from the Arctic basins carry elevated U-234/(238)Uar when compared with deep ocean seawater. Samples from the Canada Basin have a significant freshwater component and provide evidence that the Mackenzie River loses similar to 65% of its U in the Mackenzie shelf/estuary zone before entering the deeper basin. This is in contrast to samples from the Makarov Basin, which provide evidence that all of the freshwater input is derived from the major Yenisey River alone, despite the proximity of the Lena and Ob Rivers. The differing behaviour of U between the Mackenzie and Yenisey Rivers is most likely a consequence of the strong binding of U to dissolved organic matter (DOC) or secondary phases in these rivers. The Yenisey River appears to transport the majority of the DOC through the shelf and into the Makarov Basin. In contrast, the Mackenzie River appears to lose a significant amount of DOC (> 50%) in the estuary/shelf zone, which may lead to loss of associated U. These findings offer a more detailed picture of the fresh riverine water flow patterns in the Arctic Ocean when compared to other geochemical proxies. The non-conservative behaviour of U in the Mackenzie River through the shelf/estuaries has important implications for U input into oceans and the total marine budget. (c) 2007 Elsevier B.V. All rights reserved.