The contamination of riverbeds by solutes derived from the surface flow has recently received increasing attention. Channel morphological features such as bed forms are important characteristics of the stream-subsurface interface and represent one control on the rate of solute delivery from the stream to the bed. Generally, larger bed forms are expected to produce greater rates of stream-subsurface exchange. However, the longitudinal dimension (wavelength) of the bed form is also important, and this effect can produce penetration patterns that may be unexpected from a visual observation of the bed surface. Experimental tests in a recirculating flume demonstrate these effects. Commonly used mathematical models do not consider the bed form geometry explicitly and depend on the availability of calibration data to derive exchange parameters for each stream reach. More detailed models that consider the effect of bed form shape are capable of simulating some of the observed experimental results. However, existing physically based models are shown to be insufficient for some bed form geometries that may occur in real streams.