The Wankama endoreic system in the region of Niamey (Niger), monitored over the period 1992–2000, is studied with r.water.fea, a physically-based, spatially-distributed rainfall–runoff model. Catchment characteristics and data, together with model principles and construction, are described in Peugeot et al. [J. Hydrol., 2003], who used the uncalibrated model as one of several investigation tools for the screening of rainfall–runoff observations. This second paper focuses on model calibration and verification, namely the methods and criteria used to that end followed by the results thereby obtained. Based on a diagnostic function that combines errors in runoff volumes and in peak discharges, calibration is performed by exploring a 3D parameter space. A resampling-based cross-validation technique is used to investigate calibration stability with respect to data sample fluctuations, and to assess the predictive capabilities of the calibrated model. The issues of parameter uncertainty, sample representativeness, and presence of influential observations, are discussed. An empirical, non-parametric method is devised to characterize parameter uncertainty and to assign intervals to volume predictions. Model verification is performed against the data from the last two seasons. Internal catchment behavior, as produced by the model, appears qualitatively consistent with field information, including a weak upper-area contribution to catchment outflow due to large runoff abstraction by the conveying hydrographic network.