Abstract
Feed forward Multilayer Perceptron (MLP) Neural Networks are universal approximators. Weight adjustment of the connectionist model is crucial to architectures that model systems behavior. This paper developed a neural network for hydrological purposes. Two architectures were developed, investigated, and tested for forecasting rainfall in the rain-fed Sectors in Sudan. A monthly architecture and a decade architecture are developed with backpropagation feedforward neural network. The two architectures are found to be efficient for forecasting rainfall in these sectors.