Abstarct: The human brain has a considerable potential for remembering information in a long time. Heretofore, several experiments in humans are done for long-term memory retrieval. This subject can be related to retrieval of recollection or a dependent recollection. The goal of this thesis is decryption of long-term memory using by recorded brain signals. One the main brain signals is MEG. Hence, an international challenge for human memory retrieval is done in 2012. The proposed method of this thesis is tested on the databaxse of this challenge. The proposed method of this thesis has five main steps. 1) Normal feature extraction from MEG signals in time domain; 2) Feature reduction by Principle Component Analysis(PCA); 3) Premier feature selection using Genetic Algorithm (GA); 4) Mapping principle features to the new feature space by Linear Discriminant Analysis(LDA) and learning the support vector machine(SVM); finally, 5) classification of test data. Proposed method is applied on BIOMAG2012’s data. Our results show that the accuracy of proposed method is better than BIAMAG2012’s result. The next challenge is done in 2016, but the results are not published.