Aiming at solving the problem of unstable predicted results and poor generalization ability when a single Extreme Learning Machine (ELM) was treated as a classifier in the research of automatic epileptic ElectroEncephaloGram (EEG) signals classification, a classification method of AdaBoost ELM based on Mutual Information (MI) was put forward. The algorithm embedded the MI variable selection into AdaBoost ELM, regarded the final performance of the strong leaner as evaluation index, and realized the optimization of input variables and network model. Wavelet Transform (WT) was used to extract the feature of EEG signal, and the proposed classification algorithm was used to classify the UCI EEG datasets and epileptic EEG datasets of the University of Bonn. The experimental results show that compared to traditional methods and other similar studies, the proposed method significantly has improvement in the classification accuracy and stability, and has better generalization performance.