In this paper, a self-generating method based on genetic algorithms is proposed to automatically construct a fuzzy system. A parameter structure of fuzzy system is presented along with a fuzzy set elimination method, which enables the self-generating method not only to determine an appropriate fuzzy system with a high performance but also to favor the selected system with fewer fuzzy rules. An inverted pendulum control and a system modeling problem are utilized respective to illustrate the efficiency of the proposed method in the fuzzy system design.