The equivalent sound level L[inf eq] for road traffic noise is estimated using two models. One is a nonlinear neural model in which the transformation sometimes becomes analytical when it is expanded to a series. Therefore, the output function of the net is examined, and a means of partitioning the data for learning and evaluation is described. The other model is a prediction equation in which the decay due to distance is added to the unit pattern based on an expression derived from Poisson traffic flow. The variables included in the equation are then selected as inputs to the net, and learning is evaluated. The output of the net is compared with the results of the prediction equation. The neural net learns functional relations with relatively little polarization, and predictions are easily made since the data are partitioned. The degree of prediction errors is approximately the same for both the equation and the neural net.