Three-dimensional laser milling is based on the single layer milling. So, the depth of single milling determines the quality of the three-dimensional laser milling. There are many mutual factors affecting the laser milling, which can not be expressed as a linear relationship. Based on the artificial neural networks, the bidirectional feed-forward back propagation (BP) neural network models of milling width and depth are set up with Matlab platform. The neural network models are tested using samples obtained from the laser milling experiments. The results show the feed-forward BP neural network with hidden layer can predict and optimize the process parameters. Which is helpful to reduce the times of experiments.