Operations comprising a sequence of single passes whereby relative motion occurs between a workpiece and a shaping tool on each pass is termed a multipass process. This paper describes the development of a neural network modelling approach for the representation of the complex dynamic interactions that are characteristic of multipass processes. The developments are then applied to a world-scale steel beam rolling mill for the prediction of motor torque and rolling force. Two neural network structures were designed to satisfy different operational requirements. The first was to provide online single pass ahead predictions, while the second was for off-line multipass ahead predictions. Although the results obtained using the ‘best’ single network model were promising, significant prediction improvements were achieved by combining (stacking) multiple neural networks that were trained using different network topologies.