Abstract: In order to meet the need of width control precision for a 1580 hot strip mill and to improve the applicability of width spread models,the process of edge rolling-horizontal rolling in the rough rolling area was simulated by the finite element software ANSYS/LS-DYNA.The effects of width,thickness,roll diameter,reduction in width and reduction in thickness on ‘dog-bone’ width spread,natural width spread and absolute width spread were studied.The FES models of ‘dog-bone’ width spread and natural width spread were established with simulating data and field data,and the PSO-BP neural network was founded.At last,the FES width spread models were combined with the PSO-BP neural network to predict the width spread of the first,third and fifth pass.The percentage of the case that the error between predicted data and the measured values is less than 1 mm was more than 99%,and the width control precision was satisfied.