Knowledge of winter streamflow regimes is required in northern catchments to evaluate water supply and to assess the vulnerability of aquatic habitat. The objective of this study was to explore the nature and causes of winter streamflow variability in northern rivers through examination of a limited number of case studies involving intensive field measurements, as well as a synoptic analysis of winter streamflow measurements archived by Water Survey of Canada for rivers in Yukon Territory, Canada. Evidence was found for an abrupt decrease in discharge at freeze-up in one of the case studies and for 10 of the 25 stations in the synoptic analysis that had measurements within 30 days of freeze-up (an additional 12 stations had no measurements within 30 days of freeze-up). However, given the paucity of measurements in the early winter, the magnitude, duration and frequency of these events cannot be specified. The case studies indicate that, even where a coherent depression does not occur, discharge can fluctuate around a smooth recession trend for about the first 30 days after the onset of ice effects, probably as a result of transient storage and release of water behind ice jams. A storage-depletion model that represents streamflow as outflow from two parallel linear reservoirs provided a reasonable fit to most of the observed measurements (excluding those in the first 30 days following freeze-up), with model fit deteriorating with increasing latitude and decreasing catchment size. The effect of latitude could relate to abstraction of flow by ice production, which would cause deviations from a storage-depletion trend. Northern catchments also tended to have steeper late-winter recessions, which could reflect a lack of extensive, deep aquifers to maintain late-winter discharge. The tendency to poorer model fit in smaller catchments could reflect a problem with data reliability, since it is more difficult to find good winter gauging sections in smaller streams. Some evidence for temperature-related discharge fluctuations was found in both the case studies and synoptic analyses. However, the magnitude of these effects appears to be about ±10 to 15%, at most, and not to be consistent between winters. Further advances in understanding winter streamflow variability will require frequent measurements on a range of streams over a number of winters. Copyright © 2002 John Wiley & Sons, Ltd.