Abstract: Investigates the width prediction and setup of strip in rolling passes in a hot strip mill. It is found that width variation after the finishing rolling is determined by factors like the chemical composition, edging draft, longitudinal ratio, strip temperature, strip speed, strip tension, and others. According to this the width prediction during the strip rolling is classified into two processes: the dog bone rolling and the succeding finishing rolling. The mathematical model is used for the former and the principal component analysis-radial basis function (PCA-RBF) networks for the latter. It is shown that a well-trained neural network model can improve the accuracy in width prediction. Based on the prediction model, width variations in the final strips are decreased.