The assessment of water resources in a region usually must cope with a general lack of data, both in time (short observed series) as well as in space (ungauged basins). Such a lack of data is generally overcome by combining rainfall-runoff models with regionalization techniques in order to transfer information to sites without or with short available observed series. The present paper aims to analyze applicability and limitations of two regionalization procedures for estimating the parameters of simple rainfall-runoff models respectively based on a "two-step" and on a "one-step" approach, for the estimation of monthly streamflow series in ungauged basins. In particular, an application to a Sicilian river basin of multiple regression equations according to a "two-step" and a "one-step" approaches and of a "one-step" approach based on neural networks is reported. For the investigated region, results indicate that models based on the "one-step" approach appear to be robust and adequate for estimating the streamflows in ungauged basins.