In this paper, a method based on genetic algorithms is proposed to automatically extract fuzzy rules to identify a system where only its input-output data are available. This method can determine a fuzzy system with fewer fuzzy rules as well as the antecedent and consequent parameters of the fuzzy rules at the same time. A nonlinear system is utilized to illustrate the efficiency of the proposed method in the rule extraction for fuzzy modeling. (C) 1997 Elsevier Science B.V.