In order to control forming accuracy of laser cladding thin-wall metal parts, on the basis of analyzing technology theory and affecting factors of metal parts wall thickness in laser cladding, a nonlinear mapping between wall thickness of metal parts and laser power, spot diameter, scanning speed, powder feed rate is established by using BP artificial neural network. By using momentum coefficient, adaptive learn rates and optimized weights and bias, the problem of BP that easily falls into local minimum point is overcome. Experimental parameters are chosen as training samples, and BP neutral network is trained. Experimental and simulated results show that the maximum relative error of training and testing samples are 1.93% and 1.19% respectively. The optimized BP neutral network model can be used to process parameter optimization and predict forming accuracy of laser cladding metal thin-walled component for on-line control system.