This paper deals with an alternative to the classical regionalization approach, used to address the problem of ungauged basins with rainfall-runoff models. We chose a new, more practical orientation, which consists of making the most of a few discharge measurements. Indeed, it is not unrealistic to obtain such measurements by sending a gauging team on the spot where a hydrological prediction is needed. In this paper, we aim to identify the parameters of a daily lumped rainfall-runoff model, GR4J, and we look for a strategy to calibrate its parameters using a few streamflow measurements, which are combined with a priori knowledge of the parameters. Results show this approach to be much more efficient than classical regionalization studies, as soon as about thirty measurements can be made, at random, during a period of 3 to 5 years.