Abstract: In this paper, A large number of production line data was collectd and collated on welding technology and mechanical properties of welded joints. A relevant database was established based on these data, then BP neural networks optimized by genetic algorithm was used to mock up these mechanical properties of welding joint for prediction. By which mechanical properties of tensile strength, yield strength and elongation of carbon steel, low alloy high strength steel and stainless steel can be predicted. The results show that the material composition and welding technology as the main parameters of joints can be determined by this system and staisfied prediction accuracy can be obtained as well.