In order to solve the problems that gear fault classification model has weak generalization ability,poor accuracy causing by the fault features of gear is difficult to extract and extreme learning machine input weights and threshold of hidden layer nodes randomly selected,a BA- ELM gear fault diagnosis method is puts forward based on energy feature of wavelet packet optimal nodes.First,the gear vibration signals are decomposed by using wavelet packet in this method,the optimal nodes is selected by using the correlation coefficient between each node decomposition signals and original signal,and the energy feature is calculated.Second,the bat algorithm is used to optimize the extreme learning machine input weights and threshold of hidden layer node and the gear fault classification model of BA-ELM is established.Finally,the energy entropy feature vectors of the optimal wavelet packet nodes as the model input is used to identify the different fault states of gear.The experimental results show that,comparing with SVM and ELM fault classification method,the BA-ELM gear fault diagnosis method based on energy feature of wavelet packet optimal nodes has higher classification accuracy and better generalization ability.