Abstract: It has been indisputable that the high-precision gauge control should be acquired via steady on-line modification of pass or roll gap. The application of extended Kalman filter to pass modification indicated that it is able to not only solve the feedback data-processing problem, but also provide a more accurate prediction model for every updating of pass reduction. The prediction model can predict accurately the key parameters in rolling process by the linearized derivation of main models of rolling process, thus calculating the modifying values of subsequent passes with those key paramenters. The on-line application results in ANSTEEL indicated that the gauging rate increases by 14.2% for heavy paltes.