In view of the low accuracy of traditional prediction method of students’ English writing scores, a prediction model based on principal component analysis (PCA) and BP neural network was proposed. First, the dimensions of the evaluation system of students’ writings were reduced by PCA. The first three principal components were extracted to create a new sample matrix. Then the BP neural network was trained and its generalization ability was tested. The simulation results show that the maximum relative error of prediction produced by the simple BPNN is -2.165%, while the one produced by the PCABPNN is only -0.824 2%. The PCABPNN simplifies the network structure. It also improves the training rate, prediction accuracy and generalization ability of the simple BPNN. The effectiveness of the proposed model is verified.