Abstarct: Due to the effect of emotion on human life and the importance of its recognition, a lot of methods in time and frequency domain have been proposed for emotion classification. In this thesis we have used four different feature extraction methods and implimented them on DEAP dataset. This dataset has considered the emotion in four dimensions and used music-video excerpts for stimulation of the emotion. The first feature extraction method is a kind of wavelet transform called DTCWPT which has been implimented on DEAP dataset in a former research. Again we did it to compare our results with those of the paper. We classified the data by considering two different modes. First we performed the classification baxsed on the participants, then we did it baxsed on the music-video excerpts. The classification accuracy rates in the second mode were higher than the first one. In the first and third dimension the rates were slightly better than the reported ones in the paper but in the second and fourth dimension they were slightly less than the results of the paper. With the goal of improving the results we have left the frequency domain and focused on time domain features of EEG signal. For this purpose we have used CSP analysis as our feature extarction method and implimented it on the signals in two mentioned modes. This method has better results in the first mode, however its results are lower than the results of DTCWPT. In the next phase by combining the features of the two mentioned methods and by considering the first mode of the classification (baxsed on the participants) the classification rate in the second and third dimension gets better in comparision with DTCWPT approach, and it improves in the second dimension compared with CSP analysis. KNMF algorithm is the third method that we have used for feature extraction. This method is implimented on the time-frequency features extracted after the implimentation of DTCWPT on EEG signals. By the use of KNMF a new group of features is produced. However this approach singly doesn't have better results than the previous methods but if we combine the results of this one with the results of the two former methods, the accuracy rates in the second and third dimension will improve. In this study KSVD algorithm is applied too, however the classification rates are low. Among the applied methods KNMF and KSVD algorithms are so slow but CSP analysis is less time consuming and also by the use of few features has acceptable results.