Rainfall and potential evapotranspiration are the main input variables used by rainfallrunoff models to calculate river discharges at a catchment outlet. When the studied catchment is large or well monitored, gauging stations may exist upstream of the target point. Upstream flow observations may provide valuable information for downstream flow modelling and a way to account for this additional information should be proposed. Many authors report significant improvements of downstream simulations when upstream discharges are accounted for in the modelling scheme. However, streamflow data are scarce and flow values are often regarded like an output of hydrological models rather than an input. Our work has two main objectives. First, we want to confirm and explain the performance gains obtained with rainfall-runoff models that make an explicit use of upstream discharge. The generality of our conclusions is supported by a comparison of models on a large set of catchments showing a wide variety of configurations. Second, we investigate the impact of degraded upstream data on downstream simulations to challenge the robustness of the approach.