One of the still unsolved problems of Rainfall-Runoff (RR) modelling is that of the accounting for underground, discrete or diffuse (non-point source) gains or losses which affect the catchment-scale water balance. Once we have closed hydrologic textbooks and their nicely simplified diagrams, we have to face the fact that surface catchments lying on a continuous impervious horizon are the exception rather than the rule. In the same time, hydrogeologists encounter the greatest difficulties to estimate recharge, which is an essential boundary condition of groundwater systems, and often become angry at the few conceptual surface models they have to use for that task and that are not really designed for it. On this topic, Beven [2001] argued that "we cannot currently close the water balance by measurement. Traditionally, there was no direct way of measuring actual evapotranspiration, so errors in the long term measured water balances tended to be assigned to the evapotranspiration term, despite the fact that we know that rainfall inputs, discharge outputs and changes of storage are not always accurately measured. [. . . ] There is still no way of checking whether the catchment is indeed watertight. The continuity equation is the most fundamental law in hydrology, but as a hypothesis it would appear that we cannot currently verify it at the catchment scale." The aim of this communication is to present a unique case of karstic catchment, the Touvre (Charente, France), where we can test the validity of estimates of catchment leakages: this spring is fed by the losses of three rivers, which are themselves gaged before and after they reach the karst. This setting is exceptional in that it allows confronting simulations of karstic losses with actual spring discharge measurements, and thus to validate (or at least evaluate) estimates of karstic losses from surface catchments. Of course the success obtained here should not be interpreted as a definitive demonstration of the validity of the intercatchment groundwater flow estimates provided by the model we used (the daily GR4J model, see Perrin, 2003). Many more of similar cases would be needed, but at least this study recalls the importance of the rainfall-runoff information as a way to estimate groundwater recharge, not only its fraction in annual water balance but also its dynamics.