In general, vibration signals generated by the switching operation of a power distribution switch contains important information to reflect its mechanical status. A novel type of fault diagnosis method for a power distribution switch based on vibration signals analyses is proposed in this study. Firstly, the Choi-Williams distribution (CWD) for the vibration signal is calculated to obtain a two-dimensional time-frequency matrix. Then, the block singular value decomposition (BSVD) is performed on the two-dimensional time-frequency matrix, which is used to characterize the time-frequency characteristics of different mechanical states. Finally, the extreme learning machine (ELM) classification algorithm is used to train and test the feature vectors of four mechanical states of measured vibration signals. The advantages of the proposed method are that the time domain and frequency domain characteristics of vibration signals inside the power distribution switch are effectively extracted, and the diagnostic model can be trained without many samples. Experiments based on measured data show that the proposed method has a higher recognition accuracy with a faster convergent speed.