Based on artificial neural network (ANN), prediction model of the Charpy impact toughness for automatic welding is built. The input parameters of the model consist of the chemical elements and the diameter of the welding material and the outputs is the average Charpy impact toughness. The ANNs model is established by Visual C++ based on improved back-propagation (BP) arithmetic with momentum coefficients, in which the sample data used are from automatic welding materials for X70 pipeline steel. Based on the prediction model, the influence of chemical compositions, such as C, S, P, Si, Mn, Cu, Ti and Ni on the Charpy impact toughness of welding materials are analyzed. The results show that the influence of metallic elements is significantly greater than the nonmetallic, and the contents of Mn in metallic and the C in nonmetallic have primary effect on the average Charpy impact toughness.