A physically based filter for separating base flow from streamflow time series is tested using a simple dynamical model of streamflow generation from a hillslope. The model is a modified version of an integral water balance model developed for hillslopes and small basins. It does not include Hortonian overland flow, but it represents the physics of runoff generation from a hillslope better than the physics used in developing the filter. Rainfall time series data are used to run the model. The filter is applied to streamflow simulated from the model and filter estimates of base flow are compared to model-simulated base flow. The filter performs well on average, but base flow estimates include a strong high-frequency component that is not found in simulated base flow. The filter assumption that overland flow and recharge are constant fractions of rainfall is found to be a main source of estimation error. A new filter is developed using the modified model in which recharge is a delayed and attenuated response to rainfall because it is a function of integrated soil moisture. Comparing this new filter with the original one shows that base flow estimates only marginally improve, based on one criterion, for streamflow data from a first-order basin. However, this new filter provides time series estimates of integrated soil moisture and estimates of soil moisture are similar to observed trends in the basin. The formulation of this filter provides a new framework for developing better filters for estimating time series of base flow and integrated soil moisture.