Brain Machine Interface (BMI) is a fast growing technology, in which researchers aim to build a direct channel between human and machine. Data classification and the augmented time for learning and testing are common and important problems in BMI research. To overcome these problems, data normalization has been used in these systems. In this research four types of feature extraction techniques were used using the normalized and non-normalized sets of data to compare. The four techniques are Wavelet Transform (WT), Fast Fourier Transform (FFT), Principal Component Analysis (PCA), and Auto Regression (AR). In our experiment, electroencephalography (EEG) signals were extracted from one subject during three mental tasks (close arm, open arm and close hand). Data were recorded using Emotive Epoc device from four channels, AF3, F7, F3, and FC5. The classifications of the three considered tasks were done using Multi-layer Perceptron Neural Network trained by a standard back propagation algorithm (MLP-NN).The simulation results showed that, the normalization procedure enhanced the performance and increased the accuracy of the classification than the non-normalized data. Furthermore, the results showed that, WT technique with data normalization outperformed the other three methods of features extraction with classification rate reached to 92.2%.