Hot strip rolling process includes four main stages, which are reheating process, roughing and finishing process, laminar-cooling process, and coiling process respectively. Temperature is the most sensitive parameter and has direct effect on the microstructural evolution and further the mechanical properties, and the accurate control of temperature guarantees the quality of products and homogeneity of properties along the strip length. However, for the conventional hot strip rolling process, thermal history along the strip length is very complex, the related temperature variation concerns air cooling, water cooling, heat transmission by roll contact, heat generation by deformation and friction. Based on the actual hot strip mill, the thermal models are established in this paper to simulate the temperature distribution along the whole strip length from the reheating furnace exit to the down coiler. Different interface heat transmission coefficients are selected for the scale breaking and spray water-cooling process, and a self-learning algorithm is thus employed to improve the calculation accuracy. This model is characterized as simple and fast, and convenient for on-Iine/off-Iine prediction of temperature. Finally the simulated results are verified by the on-line temperature detection at typical points such as roughing exit (RT2), finishing exit (FT7) and coiling position (CT).