Abstract: Based on CMAC (cerebellar model articulation controller) neural network, a pattern recognition method for strip flatness is proposed with a flatness fuzzy controller based on the recognized results designed. Thus, the pattern recognition and controller design are combined into one. The CMAC is used to recognize the membership levels in regard to six basic patterns of common defects in flatness and then, as the direct forepiece of fuzzy controller, serve for seeking these membership levels. Analyzing the characteristics of defect in flatness, the fuzzy set is defined rationally to reduce greatly the calculation of fuzzy reasoning. The simulation result showed that the pattern recognition method of flatness offers high recognizing precision with which the designed fuzzy controller for flatness can control a defect to an expected extent with satisfactory controllability.