Researches on structural design of a vehicle for weight reduction have been steadily conducted to reduce manufacturing cost and enhance fuel efficiency.Among many parts of vehicle, the body in white (BIW) is remarkable target for weight reduction.Because it is a major part which take about 20~30% weight of a full vehicle, so weight reduction for the BIW is relatively more effective than for other parts of the vehicle.Weight reduction for the vehicle BIW is an important issue of vehicle design, but structural safety of the vehicle is also important and both two issues should be accounted for the design of the BIW.Therefore, additional to the weight of the BIW, bending and torsional stiffnesses are considered as performance responses of the design problem.In this research, thicknesses of all members of the BIW are taken as design variables of the problem.When a design object is large scale structure such as the vehicle BIW, the most difficult characteristic is that it has too large number of design variables.The BIW in this research is composed of 196 members.If we use all of the members of the BIW as design variables, it will hardly converge to optimum solution because of large number of design variables.Therefore, we summarized the 196 number of members to 64 groups according to the experts opinion and used 64 groups of the members as design variables.Although the number of design variables is reduced from 196 to 64, it is still too large to acquire optimum design with reasonable numerical cost.To address this issue, we applied a metamodel-based design approach in this research.The metamodel-based design approach is composed of a design of experiments (DOE) and metamodeling techniques, and it makes metamodels which can replace a numerically expensive real experiments or simulations and uses them for design optimization.The metamodel-based design optimization approach has several advantages;1) can gain an optimum solution more effectively, with less numerical cost, because it uses a metamodel instead of expensive real simulations, 2) can predict the time to acquire an optimum solution or can keep a given design schedule because time-consuming simulations are only taken in the DOE, and the number of samples used for DOE is determined by design engineer, 3) can investigate numerical properties of performance responses such as degree of nonlinearity or existence of correlation between them.Since there is a trade-off between the weight and stiffness of a structure, the stiffness should be sacrificed to less the weight.From the results of the metamodel-based design optimization, however, the weight of the BIW is reduced by 6.7% without the sacrifice of the stiffness.Moreover, the torsional stiffness which is more important performance is increased by 7.8% compared with the initial design.From the results of this research, the metamodel-based design optimization approach is proven to be effective for large scale design problem.