Abstract: The health condition of hot-rolling back-up rolls plays a key role in controlling the strip profile quality and rolling stability. The characteristics of nonlinearity, strong coupling, and the use of limited samples complicate the prediction of the back-up roll health state. The current back-up roll replacement strategy of each steel mill is generally determined according to a certain rolling time or rolling kilometer, and such a maintenance mode is based on experience. In actual experience, due to different strip specifications in each rolling cycle, the degrees of wear on the back-up rolls are different. Regular maintenance methods may easily lead to excessive wear of the back-up rolls and reduce the quality of the strip shape at the end of the unit, or premature roll replacement wastes the back-up roll performance. This paper proposed a construction method for the back-up roll virtual health index and a Copula function–based model for predicting the health condition of complex working conditions. The health condition of a pair of back-up rolls was characterized by combining roll bending and shifting data, and the back-up roll condition was divided by the K-means clustering method. The Copula prediction model was constructed using the process data under each working condition, and finally, according to the actual rolling schedule, the arrangement order combines the prediction results of the working conditions. The production performance data of a 1780-mm hot rolling line were used to verify the results. The results show that the proposed Copula-based prediction model can accurately and effectively predict the health condition of the back-up roll according to the rolling schedule; thus, it can serve as the basis of a more scientific strategy to guide the replacement and maintenance of the back-up roll.