As there are many parameters in the design of dual throat fluidic thrust vectoring nozzle, a method of multivariate optimization ability is investigated, which includes nozzle parametric design, using uniform experimental design (UED) to spread samples in the design range and obtaining corresponding performance, training radial basis function (RBF) neural networks with samples, and then using particle swarm optimization (PSO) to find a better nozzle parameter combination based on trained neural networks. Computational results show that the thrust-vectoring angle of the optimized nozzle is improved obviously. Tests indicate that the method is of good optimization ability, and can be used on multivariate optimization in nozzle geometry design.