The sales volume of a beverage distributor was predicted by Artificial Neural Network (RNA) in the face of the problem of over-storage in high-demand seasons. The prediction was performed for four most representative drinks of the brand: RNA-A, B, C and D with training algorithms Backpropagation (BP) and weight adjustment Levenberg-Marquadt (LM), topology: 4 inputs, population variation and time), an output (sales prediction), input transfer function the hyperbolic sigmoidal tangent (tansig) and the linear output transfer function (purelin) in 1 hidden layer, moment coefficient 0.1, target of error 0.1, learning rate 0.001. For each drink we worked with 150 neurons, choosing only 3 with the lowest mean square error "mse". RNA-A with 37 neurons, 400 training stages, mse = 0.258 and R = 0.97565 showed an absolute error rate = 9.5395. RNA-B with 90 neurons, 150 training stages, mse = 5.33 a...