An Adaptive Logic Network, a simplified, special case of the multilayer perceptron feedforward neural network, is used to model the off-line evaluation of the quality of a ball steel production line providing grinding media for the mining industry. The model developed in this project predicts the proportion of rejected bars in a cast based on significant information recorded during the process. The results of this application suggest that this method may be a faster way to develop a usable model for off-line analysis than conventional methods.