Abstract: Based on optimized BP algorithms, some prediction models are developed for cement strength. The model 1 is used to predict compression strength, of which the input is 3 d compression strength and the output is 28 d compression strength. The model 2 is used to predict bending strength, of which the input is 3 d bending strength and the output is 28 d bending strength. The BP algorithms covering additional momentum and adaptive learning rate taken to optimize the proposed BP network, which thus works efficiently. The mean values of relative errors of model 1 and model 2 are 1.6655% and 3.8341%, respectively, and the results of the prediction are proved close to the experimental ones.