Abstract: According to soil corrosion data and by means of principle of BP artificial neural network, an network simulation model for soil physicochemical property and corrosion rate of carbon steel is designed with MATLAB Neural Network Tool and a practical case shows that simulation effectiveness is remarkable and errors between predicted value and true value are minor. Prediction on corrosion rate of X70 pipe steel in soil is made. Result show that the BP artificial neural network is practical in prediction of soil corrosion rate.