A design optimization tool has been developed for the crash victim simulation software MADYMO. The crash worthiness optimization problem is characterized by a noisy behaviour of objective function and constraints. Additionally, objective function and constraint values follow from a computationally expensive numerical analysis. Sequential approximate optimization is used to deal with both the noisy functional behaviour and the high computational costs. By means of multipoint approximations, a sequence of linear programming problems is generated that can be easily solved. The optimization approach is illustrated for an analytical test problem and an industrial crash worthiness design problem.