Classical time-frequency analysis method meets great troubles while extracting efficient fault features from vibration signal of circuit breaker. This paper takes advantage of phase space transformation to extract effective features for mechanical fault diagnosis of circuit breaker by the reconstruction of one-dimensional vibration observation sequence. Firstly, it calculates the embedding dimension by G-P algorithm and delay time by mutual information method to reconstruct the phase space of vibration signals. Then it calculates the Phase Space Distribution Coefficient (PPDC) and the signal’s grid dimension of the new obtained vibration model respectively. In order to enhance the discrimination of the description of vibration signal, a novel fault feature description vector is constructed by combining the commonly used time-domain feature parameters. Finally, support vector machine is employed to classify the mechanical fault, which demonstrates that the proposed method could efficiently recognize the common mechanical faults such as operating device jamming and screws loosing, so it has good application value. This work is supported by Industry-University-Research Collaborative Project of Henan Province (No. 132107000027) and Open Laboratory for Control Engineering Key Discipline of Henan Higher Education (No. KG2014-17).