This Paper shows an approach that processes CPTEC Eta model output using Neural Networks. The main benefit of using a neural network is that it can account for nonlinear relationship between predictors and predictand. The objetive is to obtain precipitation forecasts for specific locations. The test was performed on five locations in eastern São Paulo - Brazil to summer period (1997 - 2002). In this cases, 8 models variables characterizing the circulation associated to precipitation and was selected to be used as the input nodes. Precipitation data from rain gauge was ouput nodes. This papers describes the basic strucuture and transfer function of artificial neural network. In this case is using a Feed - Forward neural network and learning algorithm backpropagation. Additionally, generating results using multiple linear regression. The performance using neural network shows favorably with those generated using linear regression forecasts and numerical model precipitation prediction. This new approach indicates a potential for more accurate precipitation forecasting