Shape control in strip rolling is intractable because of multivariate,non-linearity,time-variation and coupling of the process.In this paper,a generalized predictive control algorithm based on BP neural network model was introduced for on-line controlling the strip flatness of a reversal UC mill.Some comparative experiments were done with a conventional feedback control based on linear regression model.The results clearly demonstrated the advantage of the proposed scheme. Shape control in strip rolling is intractable because of multivariate, non-linearity, time-variation and coupling of the process. This paper, a generalized predictive control algorithm based on BP neural network model was introduced for on-line controlling the strip flatness of a reversal UC mill. Home comparative experiments were done with a conventional feedback control based on linear regression model. The results clearly demonstrated the advantage of the proposed scheme.