Abstract: With welding current, wire feed speed, welding speed and substrate temperature as input variables, weld width and residual height as output variables, and the 4-12-2 structure particle swarm optimization backpropagation（PSO-BP）neural network model of melt intert-gas welding arc additive manufacturing 316L stainess steel was built with optimal particle inertia weight and learning factor in PSO algorithm. The results show that the root-mean-square error, maximum relative error and average relative error of predicted weld width obtained by PSO-BP neural network model and expected values were 0.386, 13.477% and 2.580%, and those of weld reinforcement were 0.152, 10.372% and 2.810%, respectively. Compared with BP neural network model, PSO-BP neural network model had higher prediction accuracy and stronger stability for the prediction of weld size.