Brain Computer Interface (BCI) shown an importance tool to help paralysed patients to treatmenting their disabilities of communication and mobility by sending commands to the outside using brain signals. In this paper we propose a non linear features able to analyse the EEG signals of hand movements. These features are related to Higher Order Spectra (HOS), especially the normalized version of the third order spectrum namely the bicoherence. The Support Vector Machine (SVM) is used to classify the proposed features into their pertinent classes (left or right hand movement). The SVM classification of our features achieved an average accuracy of 79,14%. We concluded that HOS study could be an accurate tool in the recognition of hand movement.