Abstract: Vibration signals of hydropower units are typically non-linear and non-stationary. To diagnose and analyze such signals, this paper presents an original signal processing method of applying empirical mode decomposition (EMD) and demonstrates the then calculation procedure of intrinsic mode functions and their complexity features. Fault diagnosis of the signals was carried out using the least squares support vector machine (LS-SVM), and by taking the radial basis function as the kernel function, its relevant parameters were determined through grid search and cross validation. The results show that coupling EMD decomposition with SVM in analysis of complexity features provides a rather accurate device for fault diagnosis and determination of fault type, thus laying a basis for operation and maintenance of hydropower units.