Abstract: A comparative study was conducted to the performance between local mean decomposition (LMD) and empirical mode decomposition (EMD) in diagnosis of simulation and experimental fault signals of turbine rotors, after which the end effect evaluation indexes and fault diagnosis method for rotor imbalance, oil whirl and rubbing phenomenon were put forward based on LMD. Meanwhile, the signal decomposition results based on EMD and LMD were compared and analyzed. Results show that by the method of LMD to decompose the rotor fault signals, the level of mode mixing is lower than that of EMD. Also, LMD has a stronger capability in inhibiting end effect and a higher precision in fault diagnosis, compared with EMD algorithm, which therefore can be effectively applied in fault diagnosis of turbine rotors.