Hydraulic engineering plays an important role in energy construction in China. As the most important water retaining structure, the deformation trend and safety state of dam is undoubtedly the most concerned problem in engineering. Dam deformation monitoring data is the most critical information to understand dam deformation. So, the analysis and prediction of dam deformation monitoring data is an important measure to master dam safety state. However, the monitoring data of dam generally contains noise components. In order to reduce the noise influence and improve the stability and accuracy of dam monitoring data. EMD-SARIMA model was established in this paper. The monitoring data was decomposed into several Intrinsic Mode Function (IMF) from high to low frequency by using Empirical Mode Decomposition (EMD). Then, the data was reconstructed after eliminating the IMF mainly containing noise based on the Continuous Mean Square Error (CMSE) criterion. Finally, a Seasonal Auto-Regressive Integrated Moving Average (SARIMA) model was established for the reconstructed data. The results show that EMD can effectively reduce the noise in dam monitoring data. The reconstructed data is more stable than the original data, and closer to the actual displacement process of the dam. Compared with SARIMA model, the prediction accuracy of EMD-SARIMA model meets the requirements, and is more accurate and less noise effect. It can be applied to denoise data and prediction analysis of gravity dam.