To solve the problem of low classification accuracy due to incorrect feature extraction in the existing fault diagnosis algorithms for high voltage circuit breaker, the fault recognition algorithm based on deep belief network is proposed. Deep Belief Network (DBN), as an unsupervised deep learning algorithm, is composed of multiple Restricted Boltzmann Machines (RBM). Firstly, using the unlabeled data samples, each RBM layer is trained layer by layer from bottom to top, and the optimal parameters of each layer are obtained. Secondly, using the parameters obtained before as the initial parameters, the DBN is expanded into a structure that propagates backwards. Then the labeled data samples are used for global parameter tuning. Finally, the DBN classification network is established. In this process, the artificial feature extraction is effectively avoided, the local optimum problem of network training is solved, and the circuit breaker fault diagnosis is more intelligent. Experimental results show that this method can be applied to the diagnosis of major mechanical faults of circuit breakers accurately and reliably.