A central composite rotatable experimental design (CCRD) is conducted to design experiments of laser transmission welding of thermoplastic-polycarbonate (PC). The genetic algorithm-artificial neural network (GA-ANN) and response surface methodology (RSM) models which establish the relationships of the laser transmission welding process parameter (laser power, scanning speed, clamping pressure, scan times) and joint strength are established, and then the welding strength is predicted and the welding parameters is optimized by using the developed models respectively. The modeling capabilities, generalization and optimization capabilities of the two models are systematically compared. The result shows that GA-ANN and RSM are not significantly different on the maximum experimental joint strength, but the modeling, generalization and optimization abilities of GA-ANN are better than that of RSM, so GA-ANN is a more effective way to optimize the PC joint strength.