Most often reactor designs are scaled-up using experimental measurements; especially for the manufacture of fine and specialty chemicals. Yet without an appropriate model of the reaction system major opportunities can be missed in the design and optimization of the reactor. In this paper a model building methodology for complex reaction systems combined with experimental evaluation is developed. Preliminary experiments provide information for construction of a reaction network containing all feasible reactions. A strategy is employed for generating rival alternative simple model's using different subsets (mechanisms or pathways) extracted from the reaction network. On the basis of model predictions further experiments can be carried out either at the optimal reactor performance conditions or at conditions that allow discrimination among rival models. These additional experiments can be used to refit and improve the rival models and to test and validate the predicted optimal reactor design. An example involving oleic acid epoxidation is used to demonstrate the methodology.