A new method for fault diagnosis of high voltage circuit breakers is proposed based on empirical wavelet transform (EWT) and multi-scale entropy. Firstly, the original vibration signals of high voltage circuit breakers are decomposed into a number of intrinsic mode functions by EWT method, reconstructing the signal with intrinsic mode function (IMF) whose correlation coefficient is bigger than others. Secondly, the vector that stands for circuit breaker working condition is extracted from reconstructed signals based on the mult-scale entropy. The feature vector is preprocessing with the method of normalization considered as the input vector of support vector machine. Lastly, based on grid algorithm optimization of its kernel functions, the support vector machine can classify the different states of the circuit breaker after importing the feature vector of test sample. The experimental results indicate that the method can fast and accurately diagnose breaker faults, and identify the states of circuit breaker.