Abstract: Discussing the existing empirical formulae of roll wear due to plate rolling, an on-line roll wear model based on regression analysis was studied utilizing the wear-out data acquired in situ. Based on the statistical wear-out distribution along roll barrel, the wearing value of a roll is considered as a function of roll length to predict the roll wear in next rolling period. Predictions by relevant model based on statistical regression analysis and measurement were both carried out for roll wear, with a roll taken from the 3500 mm plate mill in Shougang Steel Plant as specimen. The comparison shows that the predicted values fit with measured values well. The roll wear prediction model can improve significantly the precision of calculated values of roll wear obtained from empirical formula, this providing favorable controls of thickness precision, shape and crowning of plates.