Most of the popular EEG classifiers need to be supervised and their parameters have to be trained by a number of train data in advance.That’s the reason why they cannot be used in the real-time circumstances.In this paper，a new FCM unsupervised classification algorithm is proposed which is based on the density size of data dot and mahalanobis distance.Then，the algorithm is used to classify the EEG signals from the database of the second session of 2003 BCI competition.The EMD algorithm is used to decompose the EEG and extract the characteristic values，and then these values are classified by the proposed FCM algorithm.The experimental results show the algorithm’s feasibility and validity in the EEG classification field.