In order to improve the lightweight level and performance of the car door, the integrated method of "material-structure-performance" is adopted to design the cast aluminum integrated car door. Based on the constructed finite element model of the cast aluminum integrated car door, with the thickness of the car door as the design variable, the radial basis function (RBF) neural network approximation model and the second-order response surface approximation model were used in combination with the non-dominated sorting genetic algorithm-Ⅱ (NSGA-Ⅱ), multi-objective particle swarm optimization (MOPSO) algorithm and multi-island genetic algorithm (MIGA) to conduct the deterministic optimization design for the sinking stiffness condition displacement, upper torsional stiffness condition displacement, lower torsional stiffness condition displacement, first-order bending mode frequency, first-order torsional mode frequency and mass of the car door. On this basis, the 6Sigma reliability analysis and optimization for the quality level of the deterministic optimization solution were carried out considering the uncertainties of materials and manufacturing. The results showed the optimal combination of second-order response surface approximation model and MOPSO algorithm achieved the optimal lightweight of the car door, and the optimal combination of RBF neural network approximation model and MOPSO algorithm could minimize the the displacement of the car door under the sinking stiffness condition. The above two combinations achieved the design goals of lightweight and safety for the car door, respectively. The research results can provide reference for the optimization design of car parts.