The potential of artificial neural network models for simulating the hydrologic behaviour of catchments is presented in this paper. The main purpose has been the modelling of river flow in a multi-gauging station catchment and real time prediction of peak flow downstream. The study area covers the Upper Derwent River catchment located in River Trent basin. The river flow has been predicted using upstream measured data. Three types of ANN were used for this application: Multi-layer perceptron, Recurrent and Lagged time recurrent neural networks. Data of different lengths (1 month, 6 months and 3 years) have been used, and flow with 3, 6, 9 and 12 hours lead-time has been predicted. In general, although the ANN shows a good capability to model river flow and predict downstream discharge by using only upstream flow data, however the type of ANN as well as the characteristics of the training data were found very important factors affecting the efficiency of the results

Keywords