Understanding stream runoff generation processes requires distributed stream runoff estimates; however the acquisition of such estimates remains challenging in hydrology, especially in remote areas. In regions with a high spatial density of small reservoirs, those reservoirs could be employed to gauge stream runoff (Liebe et al., 2009). Using a water balance approach, the stream runoff flowing into a reservoir from a drainage catchment could be estimated. Accordingly, this work aims to address the following two questions: i) what is the error in the estimated stream runoff and ii) what are the main estimation uncertainty factors? Based on a case study of the Kamech catchment, Tunisia, stream runoff was estimated at different temporal resolutions (1–32 days), and a global sensitivity analysis was performed to estimate the contributions of the reservoir water balance terms (evaporation, rainfall, percolation, reservoir water level and level-area-volume relations) to the estimated stream runoff uncertainty. The results reveal that stream runoff can be reliably estimated based on small reservoirs using a mass balance approach. The error and global stream runoff estimation uncertainties decrease as the temporal resolution increases. The bathymetric relationships (level-area and level-volume relations) constitute a strong factor of uncertainty for all temporal resolutions, and even the dominant factor for temporal resolutions ranging from 4 to 23 days. The estimation uncertainty for the shortest temporal resolutions (1–8 days) mainly originates from reservoir level uncertainty. As the temporal resolution increases, the contribution of percolation uncertainty increases. The general (not site-specific) conclusions of this study are that stream runoff gauging based on small reservoirs requires the determination of the bathymetric relations and that small reservoirs could be used as reliable stream runoff gauges at short temporal resolutions if the reservoir level is measured with limited uncertainty and at long temporal resolutions as long as the percolation rate from the reservoir is known with limited uncertainty