This paper presents a classification method of epilepsy electroencephalogram (EEG) signals ba sed on the combination of Hilbert marginal spectrum and extreme learning machine .Firstly,the empirical mode decomposition of EEG signals is carried out,and t he Hilbert marginal spectrum is obtained by applying the Hilbert transform to th e first 5intrinsic mode function; Then Shannon entropy,Renyi entropy and Tsall is entropy of Hilbert marginal spectrum as well as the energy of 5different fre quency sub-band rhythm signals were input into the extreme learning machine as e ffective characteristics for classification.The experimental results show that the classification accuracy of the epileptic signal in this paper reaches 99.8%, which is higher than other classification methods in detection accuracy and com puting speed,and has potential application value in real-time detection of epi leptic seizures.