In the process of hot rolling silicon steel, roll wear directly affect its shape. Accurate prediction of roll wear is an important condition for rolling qualified silicon steel strips. The traditional roll wear prediction model is established by the slicing method. The wear of F5–F7 work rolls used for finishing rolling silicon steel on a 2250 mm production line in a steel mill was predicted by this model. It was found that there was deviation between the predicted results and the actual wear, and the prediction accuracy of the model was insufficient. Therefore, the wear of the surfaces of the rolls used for rolling silicon steel on this production line was studied. Based on the analysis of the work roll wear’s form and the rolling parameters that affect the roll wear, the traditional roll wear prediction model was optimized by the genetic algorithm. Finally, the optimized model was verified, and the prediction accuracy of the wear prediction model improved. The accurate prediction results provide a basis for the formulation of a shape control strategy when rolling silicon steel on this production line.