Introduction F UZZY logiccontrolhas beenproven to be a powerful toolwhen it is applied to various control problems.1i4 In general, fuzzy logic control needs to establish fuzzy inference rules, which are preconstructedby an expert. When the rule base, which represents the experience and intuition of human experts, is not available, an ef cient control cannot be expected. To tackle this problem, selforganizing fuzzy logic controllershave been proposed. This kind of controller has a learning algorithm and is capable of generating and modifying control rules based on an evaluation of the system’s performance. The modi cation of control rules is achieved by assigning a credit to the control action based on the present performance. However, the self-organizing fuzzy control proposed in Refs. 5 and 6 has some problems. Its control rules are sensitive to set-point changes. And the learning algorithmmay generate unreliable credit value and lead to incorrect rule modi cation. Also, the convergent time of the control action is tedious because only the red rule is modi ed each time, and nally, the convergence of the control action is not guaranteed. This Note proposes a new type of design method of selforganizing fuzzy controller and illustrates its application for ight systemcontrol.The proposedmodel referenceself-organizingfuzzy controller (MR-SOFC) has two suites of fuzzy logic; one is for control and the other is for learning. The output of the referencemodel is used as a reference for rule modi cation instead of a set point, and so incorrect modi cation caused by change of set point can be avoided. Also, the learning algorithm will modify the control rules according to the fuzzy inference of the reference model output error and its derivative instead of by the xed value, so that the learning algorithm can proceed more reasonably and the learned rules can converge more quickly and accurately. The MR-SOFC can start to work even from an empty rule base and can achieve satisfactory control performance after several learning runs. By applying the proposed MR-SOFC to a ight control system, the simulations illustrate that this MR-SOFC can achieve satisfactory performanceand robustnesswhen the ight system is subjected to plant variations arising from different ight conditions.