Abstract: The temperature model of hot strip finishing mills in tandem was discussed, as well as the conventional self-learning strategy relevant to it. To rise above the defects in the conventional self-learning strategy, a zonal compensation method is applied to temperature model self-learning, where the finish rolling temperature errors are assigned to each cooling zone according to certain temperature distribution coefficients, which can be readjusted according to the rolling force of each and every stand. In such a way the predicted precision of not only the rolling force but also the finish rolling temperature can be improved well. The new temperature model self-learning strategy has successfully been applied in Tianjin RockCheck 750 mm hot strip plant.