A Fuzzy Finite State Machine (FFSM) is an extension of the classical Finite Deterministic Automaton (FDA). This algorithm finds many applications in several fields as pattern recognition, intelligent agent modeling, human - machine interfaces, and finally but not less important control engineering. This paper proposes novel strategies to approach two of the main problems in designing FFSMs: structure definition from a “seed” FDA (basic prototype), and simplification of the resulting model according to the statistical features of both external and feedback inputs.