In traditional metamodel-based robust design methods, the dependency between metamodels and true responses will inevitably introduce so-called metamodel uncertainty in robust design. Previous robust design methods often treat a metamodel as the real model and ignore the influence of metamodel uncertainty in robust solutions. It often finds a solution in the infeasible region. So the concept of robust design is extended for both parametric uncertainty and metamodel uncertainty to evaluate the compound effect in system responses. A mathematical example was employed to illustrate the superior performance of the proposed robust design method. The method was applied to an autobody lightweight design problem, in which the high-dimensional and high-nonlinear crashworthiness responses are chosen as constraint functions. The results show that the proposed method can reduce the risk of constraint violation due to metamodel uncertainty. The robust solution improves the impact performances and reduces the structure weight by 11.06%. It demonstrates the proposed method is an effective tool in autobody robust design.