According to the mutli-stage traits of manufacturing process of hot-rolling strip, a genetic neural network model with high-dimension multi-input layers is built in this paper to predict the product’s mechanical property. This model can be seen as the result of adding input node to some hidden layers of neural network according to the conducting order of technological process, so that it can simulate the manufacturing process much better. Meanwhile, In order to avoid local extreme point caused by standard BP algorithm, genetic algorithm is adopted in this paper to conduct global pretreatment for weights and thresholds of neural network. Then standard BP algorithm is used to do the training to compensate mutually both advantages and disadvantages, and thereby get global optional solution. Finally, testing results given by the actual manufacture date of hot-rolled products from an iron and steel enterprise show that the predicted result satisfies the requirement of demand. Moreover, it has higher accuracy and stability than classic BP and RBF neural network..