In order to optimize the processing parameters of plasma spray welding and improve the quality of the spray welding layer, by combining RBF neural network approximation model with NSGA-II genetic algorithm, the plasma spray welding test datas were trained based on MATLAB platform to construct the approximate model of microhardness, wear amount and dilution rate. The next multi-objective optimization of the model was carried out by using NSGA-II genetic algorithm. Finally, the pareto optimal solution set was obtained, and the interaction between process parameters was studied. Experimental results show that RBF-NSGA-II genetic algorithm can improve the quality of spray welding layer more significantly than the response surface method. It is concluded that it has certain reference value for the optimization of plasma spray welding process．