Abstract: Back-Propagation Neural Network was used to develop the gasoline blending mathematical model for a specific petroleum refinery. Based on the characteristics of its gasoline production units, the topological structure of the neural network was decided. By using real gasoline production data of the refinery, the learning function of the network and the number of neurons in the hidden layers were selected, the model was trained, and a gasoline blending model with reasonable fitting and predicting ability was established. Compared with other blending models, the model established in this paper is much more adaptive because it does not require knowing blending mechanism. Application results showed that the neural network model could precisely predict those nonlinear parameters like octane number, induction period, etc., and also provide optimized gasoline blending schemes according to real production data.