Development of artificial neural networks (ANN) for rainfall forecasting. A four stage network development procedure is adopted which involves identifying appropriate networks, determinig network complexity, estimating parameters and evaluating performance. Particular emphasis is made on the generalization issue of complex networks with large parameters. The upper Parramatta River catchment is used as a test case to compare the three alternative ANN, namely multi-layer feedforward network (MLFN), partial recurrent neural network (PRNN) and time delay neural network (TDNN). It was found that with careful development, the three alternative types of network could produce reasonable predictions of the rainfall depth one time-step (15 min) in advance. In addition, various ways to improve the accuracy of forecasts were attempted. By integrating an ANN with a spatial model developed within a Geographical Information System environment, the rainfall forecasting model was able to forecast the spatially distributed rainfall patterns one time-step ahead for the study catchment.