Abstract: Research on the application of artificial neural networks to the prediction of runoff from ungauged catchments is presented. Available catchment descriptors have been used as input data and the index flood as output. Different types and numbers of catchment descriptors were used to ascertain which gave the best relationship with the hydrological behavior and flood magnitude. Different architectures of ANN were developed and evaluated. Results show that the selection of pooling groups of catchments either randomly or according to geographical proximity does not produce desirable results. Therefore, hydrologically similar catchments were clustered using the Flood Estimation Handbook software and this improved the accuracy of the predictions. Finally, a comparison of the ANN approach and the Flood Estimation Handbook is described that shows the advantages of the ANN approach.