Abstract: The separation conditions of Cl-,NO3- and NO2- in high mass concentration of chloride are presented by uniform design.With experimental conditions and results as training collections, the architecture and learning rates of Back-Propagation Artificial Neural Network (BP-ANN) are optimized. Having been trained 2567 times, the errors of the network are in 10-5 order.The examination for the training results indicates that the artificial neural network sets up successfully the prediction models of ion chromatographic separation with less experimental data than those from general experimental method.The maximum relative error is only 0.85%.