The model of theoretical potential function represented by networks has been proposed. Quantum chemical ab initio method is used to calculate the interionic potential of molten salt systems. The physical contents of network potential are indicated with Born-Mayer-Huggins empirical potential functions, whose deviation from the theoretical potential owing to its approximation is corrected using back propagation artificial neural networks (BP-ANN). Preliminary results have shown that the theoretical potential function networks are useful for the computer simulations of molten salt systems. The model of theoretical potential function represented by networks has been proposed. Quantum chemical ab initio method is used to calculate the interionic potential of molten salt systems. The physical contents of network potential are indicated with Born-Mayer-Huggins empirical potential functions, whose deviation from the theoretical potential due to its approximation is corrected using back propagation artificial neural networks (BP-ANN). Preliminary results have shown that the theoretical potential function networks are useful for the computer simulations of molten salt systems.