Abstract: This paper proposes a new architecture of high-dimension input wavelet neural network based on key inputs and work procedures for modeling complex large-scale industrial systems. In the neural network, some input variables are connected directly to the second hidden layer or other later hidden layers according to their actions being early or late in the work procedure. The key input nodes are connected to not only all nodes in the subsequent layer but also the output node The developed method is applied to build a product quality model for continuous casting furnace and hot rolling mill. Simulation results demonstrate that the proposed network is more efficient and provides a higher accuracy compared with other four neural networks methods.