The warm and humid Equatorial characteristics of Manaus, Amazonas as well as its geographic location and the topography gives the city a naturally uncomfortable weather. Because of that rainfall prediction in Manaus has an additional importance: it improves its citizens well-being. However, predicting rainfall in Manaus is a complex problem because it is under the influence of many precipitating systems. In this work, we addressed the problem of rainfall prediction in Manaus by using multilayer artificial neural networks. The input data was obtained from an automatic weather station during the years of 1970 to 2015. The performance factor considered was the normalized mean squared error. According to the results observed, a feedforward neural network with 2 hidden layers with 10 neurons each was the one that best addressed this problem. We also could see that recurrence in the neural networks did not improve the performance in the problem under consideration.