At present the optimal design of a complex mechanical component, assembly and system is often carried out using overly simplified and less reliable mathematical models. In addition, many costly and time-consuming physical prototypes have to be developed to improve an industrial design through a lengthy trial-and-error process. In this work, a global optimal design method based on the parametric virtual prototypes of a mechanical design is proposed. The optimal design of multiple functional panels of a Proton Exchange Membrane (PEM) fuel cell stack is used to illustrate the approach. An efficient global optimization search technique, the adaptive response surface method, is applied in the design.