Multi-label Random Forest (ML-RF) is proposed and applied to the recognition of multiple power quality disturbances. ML-RF is an ensemble learning algorithm based on the Multi-label Decision Tree (ML-DT), by assembling sub-classification decision tree to enhance the overall performance. Firstly, all power quality disturbances are decomposed by steady wavelet transform, and the energy entropy of the wavelet coefficients are extracted as eigenvectors. Then, it uses training sets that are constructed by Bootstrap re-sampling method and subspace sampling method to train sub-DTs. Finally, it combines the sub-DTs by majority voting to predict the type of multiple power quality disturbances. The simulation results show that the ML-RF can effectively recognize the multiple power quality disturbances under different noise conditions, and it is a feasible method with noise robustness. This work is supported by National Natural Science Foundation of China (No. 51277134).