The main idea of our approach is to combine discrete-event simulation and exact optimization for supply chain network models. Simulation models are constructed in order to mimic a real system including all necessary stochastic and nonlinear elements. Such simulation models are used as proving grounds for analyzing and improving a real situation on a trial-and-error basis. A traditional optimization method on top of a simulation model has major disadvantages: The optimization method uses the simulation model as a black-box. Information about the structure of the problem is not available and cannot be used for an intelligent optimization strategy. On the other hand pure optimization models used for planning scenarios are usually built on a very abstract level neglecting possibly important nonlinear and stochastic properties. This is necessary, because otherwise the resulting complex optimization models cannot be solved and are therefore of no use.