Abstract: Cavitation has an important influence on the performance of centrifugal pumps, but in practice it cannot be detected by monitoring the operation state of a centrifugal pump until it has developed into a very serious case. To develop a practical technique for detecting such cavitation in advance, we should examine the behavior of cavitation inception in centrifugal pumps. This study first used a numerical simulation method to predict the cavitation inception point for a centrifugal pump, and then conducted experimental tests on cavitation noises at this inception point and collected the cavitation noise signals. Finally, empirical mode decomposition algorithm (EMD) was used to decompose the signals and find out their characteristics that describe the cavitation inception points. Results show that EMD is an effective algorithm for detection of the changes in cavitation morphology of a centrifugal pump and the cavitation inception points predicted by our numerical simulations agree well with those measured in the experiment.