With the complexity of the underground metal geological and mining conditions, the influencing factors of the rock displacement are an complicated issue and they are interactive, which leads to the great uncertainty of the prediction of the rock displacement. Large amounts of sample data slow down the training process of the neural network, which makes the neural network instable. The PCA (Principal Component Analysis) is, therefore, used, combined with the Elman network, to build the model for prediction of the underground mining rock displacement angle. The principal component analysis is used in the raw data pre-processing to extract the main ingredients from the original information, to reduce the amount of input data and make them unrelated, thus to speed up the neural network training process; the samples are then trained with the Elman network, and with the trained network to make predictions for the samples. Compared with the predicted results obtained without using the PCA, the predicted results obtained by using the PCA are more accurate. Through comparing the expected output with the actual output, the relative errors are less than 5%, which shows that the PCA combined with the Elman network is good for the prediction of the underground mining rock displacement.