Abstarct: Nowadays EEG signals have been studied by many scientists around the world. They are being used in some applications such as illness detection, determining level of consciousness and motor control subjects (such as BCI). Due to increasing number of systems baxsed on EEG signals, lack of proper databaxses in our country in this area in one hand, and the importance of these databaxses for getting good results in other hand, lead us to create a proper databaxse to control arm movement using motor imagery brain signals. For creating this databaxse, 3 volunteers were asked to imagine four specific arm movements. This databaxse includes 100 tests from 4 different arm movements in terms of eyes open and closed. Also in this thesis, Matching Pursuit method is used both to evaluate our databaxse and as a novel method for feature extraction. According to our knowledge, it has not been used for this application so far. Furthermore, this method has been compared with wavelet and wavelet packet for feature extraction where in all these methods, SVM classification method has been used. According to the simulation results, the proposed method could increase classification accuracy partially, having lower variances rather than two mentioned conventional methods. Also the results of recognition accuracy can be improved by selecting proper dictionaries and iterations. Moreover in order to implementation, at first a suitable hardware was designed and built to capture, preprocess, and convert two EEG channels from brain signals in occipital region. Some features of our hardware system are portability and low consumption power (24 mW). Then, an electronic processing system baxsed on TMS320C5509A DSP processor was designed and built too. This system is also portable and could work with batteries. Then the overal system (EEG preprocessing and processing) is trained to distinguish two separate imagery arm movement for a specified person. Results show that detection of motor imagery can be done by choosing an appropriate threshold for their coefficients energy in Beta frequency band with acceptal recognition accuracy.