Symbolic representation and sequential processing of information in most expert systems have made it difficult to represent continuous variables and to model interactions among different parts of a large process system. This paper proposes a new approach, principally aimed at function oriented engineering design problems. Continuous variables and qualitative information are treated as fuzzy sets. A fuzzy match inference strategy is proposed to assess design schemes with rules in parallel. The fuzzy match inference strategy is then extended to multi-step decision making in the synthesis of multicomponent separation sequences. The new concepts have been implemented in a prototype expert system, where the selection of separation technologies and sequencing of separators are approached in an integrated manner. It is shown from several examples that the inference strategy yields promising results in quantitative evaluation of process synthesis and combinatorial optimization.