Seeking quantitative relation between the microstructure and properties is a key point for on-line and off-line predictive system in hot rolling process. In order to develop a system for predicting properties of hot rolling strip steel SS400 during hot rolling process, a model of neural network model used in predicting properties of hot rolling strip is developed based on Matlab neural network toolbox. By using a back propagation (BP) algorithm, to eliminate the shortcoming of general back-propagation algorithm, Matlab neural network toolbox has modified BP algorithms. Using more effective numerical optimization algorithms, such as Levenberg-Marquardt, the relation between chemical composition and main hot rolling parameter and mechanical properties of product has been built in this prediction model. The analysis indicates that the model is not only simpler in programming, but also quicker in convergence. It is concluded that the main factors influencing the yield strength, tensile strength and elongation of steel are strip thickness and carbon content.