Abstract: In view of the intrinsic imperfection in identifying roll eccentricity by way of the fast Fourier transform, a method is proposed to identify the roll eccentricity, based on wavelet transform to improve the delivery gauge of hot-rolled strips, etc., the difference of transmission characteristics of the maximum wavelet modulus between singular signal and random noise in multiscale space is utilized to analyze the signals from rolling load and thickness deviation so as to separate the signal from random noise and realize the dynamic compensation for roll eccentricity. The simulation results show that the roll eccentricity compensation based on wavelet transform is better than the fast Fourier transform or its modification for improving delivery gauge of which the thickness fluctuation due to roll eccentricity can be reduced by more than 90%. The proposed method is more effective and adaptive to outside interference especially to real-time dynamic compensation control of roll eccentricity.