The paper presents a systematic approach for the extraction, interpretation and exploitation of design knowledge in process synthesis. Knowledge is developed in the course of superstructure optimisation. Semantic models (ontologies) and analytical tools are combined to simplify the superstructures and interpret solutions. In the course of the search the method translates intermediate solutions and upgrades the superstructure model. The approach supports a faster implementation and a transparent interpretation of the solutions. Results are presented for the synthesis problem of reactor networks, essentially addressing the challenges of a multi-level optimization problem. Although presented with stochastic optimization techniques, the proposed method is applicable to general types of models and optimization methods.