Due to the intrinsic nonlinear characteristics and complex structure of the high-speed catenary system, a multi-objective design optimization method is proposed based on neural network and genetic algorithm, the dropper space and the tensions of wires are selected as design variables. The multi-objective optimization involves two optimization objectives. One is the standard deviation of contact force, and the other is the maximum uplift at the support. To achieve this, an approximation model of 400 km/h railway catenary based on feedforward neural network is trained and constructed by using the sample set. The optimal value of the design parameter combination is searched by nondominated sorting genetic algorithm II. Finally, the improved Sobol method is adopted to analyze the sensitivity of these design parameters. The optimization results show that not only the standard deviation of the contact force is decreased by 19.63%, but also the maximum uplift at the support is decreased by 17.54%.