Abstract: Firstly, the BP neural network model was established by artificial neural network in this paper and neural network procedures were compiled with Matlab language. Secondly, the experimental data were trained by the network. Thirdly, the relationship between adsorption time and unit adsorption amount was simulated by trained network and the simulated results and experimental data were compared. It was concluded that the mean square error(MSE) respectively were 0.9719(Zinc ions), 0.2398(Copper ions), 0.9352(Lead ions). The results showed that forecast of network on Na-bentonite adsorbed copper ions was best, followed by forecast for lead ions adsorption, the forecast of zinc ion adsorption was worst.