Abstract: Cavitation of water turbine units is one of the factors influencing the stability and efficiency of their operation. Cavitation phenomena that occur in a running turbine unit are not easy to observe directly and acoustic signal monitoring is an effective way of the cavitation study. Traditional Fourier transform and wavelet transform are two commonly-used methods for analysis of narrow-banded low-frequency signals, but they are difficult to cover high-frequency components of water turbine cavitation broadband signals. This paper describes a new self-adaptive Hilbert-Huang transform (HHT) method that uses EMD decomposition to extract the physically meaningful signal components of turbine cavitation mode, and explores cavitation identification using HHT. Experimental tests show that this method is fast in diagnosis and better in generalization performances. Thus, it is a suitable method for cavitation fault diagnosis of hydroelectric unit.