Abstract: The cement raw materials blending system is very complex for its multi-variable, long time delay and nonlinear characters. It is difficult to set up an exact mathematic model with conventional optimum methods. Artificial neural Networks are able to give new solutions to such problems due to their capacities of nonlinear mapping, character take-out and error tolerance. According to technical requirements of cement raw materials blending, this paper applied neural network model based on BP algorithm to pre-estimate the quality of cement, to realize the optimization of cement raw materials.