A filter for separating base flow from streamflow in a river basin is derived from a mass balance equation for base flow from a hillside. Unlike other existing filters, this filter is physically based both during streamflow recession and rise, and each of the parameters in it has a well-defined physical meaning. Unlike other approaches to base flow separation, the filter is applied without any calibration or constraints. Yet, using two criteria as measures of filter performance, it performs as well as or better than two existing filters that require calibration. Using the same two criteria, weaknesses in the filter are identified and attributed mainly to a ratio of two parameters that are related to groundwater recharge and overland flow. This ratio is estimated using precipitation and streamflow data, and its estimate is conjectured to be inaccurate because of the poor spatial resolution of the precipitation data. Moreover, this ratio is assumed to be constant in time, yet, on physical grounds, it should be treated as a time-dependent quantity. ¿ 2001 American Geophysical Union