A common approach in modeling surface water-groundwater interaction is to represent the streambed as a homogeneous geological structure with hydraulic properties obtained by means of model calibration. In reality, streambeds are highly heterogeneous, and there are currently no methodical investigations to justify the simplification of this geologic complexity. Using a physically based numerical model, synthetic surface water-groundwater infiltration flux data were generated using heterogeneous streambeds for losing connected, losing transitional and losing disconnected streams. Homogeneous streambed hydraulic conductivities were calibrated to reproduce these fluxes. The homogeneous equivalents were used for predicting infiltration fluxes between streams and the aquifer under different hydrological conditions (i.e. for different states of connection). Homogeneous equivalents are shown to only accurately reproduce infiltration fluxes if both the calibration and prediction are made for a connected flow regime, or if both the calibration and prediction are made for a disconnected flow regime. The greatest errors in flux (±34%) using homogeneous equivalents occurred when there was a mismatch between the flow regime of the observation data and the prediction. These errors are comparatively small when compared with field measurement errors for hydraulic conductivity, however over long river reaches these errors can amount to significant volumes of water.