The brain-computer interface (BCI) is a communication system and direct control. It relies solely on the cerebral activity of a person and an electrical or mechanical system without muscular response. In our project, we contribute to the optimization of certain steps and particularly the features extraction of the electroencephalogram (EEG) and the classification of mental tasks in BCI. The BCI is based on the measurement and analysis of signals get crops on the scalp of an individual using surface electrodes, it operates asynchronously since the voluntary production of cerebral activity. In indeed, this activity recorded represents the motor cortex changes during motor activity at the completion of the movement or when the subject imagines a movement. From the methodological point of view, we proposed a technique for extracting characteristics for EEG signals. This method will allow us put in evidence the power in the sensorimotor rhythms when the subject imagines a movement. In addition, the same method will apply for electromyogram (EMG) for more precision. Finally, we implement with MATLAB a model that allows Online translate all of these methods to decode the mental state of the subject. Key words: Brain -Machine Interface, EEG processing, feature extraction, classification signal.