Abstarct: Brain-Computer Interface (BCI) is a useful communication tool between the human’s brain and external devices. Accurate and effective classification of Electroencephalography (EEG) signals plays an important role in performance of BCI applications. In this research, two EEG signal classification approaches baxsed on sparse representation and dictionary learning is proposed to classify mental tasks. We used Colorado and Berlin university databaxses that includes two and five classes respectively. Band pass filtering and common spatial pattern (CSP), which are effective in BCI, were applied to preprocess data. In first proposed method a classifier baxsed on sparse representation was presented. In this method feature extracted from CSP were used to build a dictionary and to apply to a multiclass problem, multiclass CSP was used. In second proposed method classification baxsed on dictionary learning was proposed. In this method an analytical dictionary (to calculate sparse response) and a synthetic dictionary (to minimize reconstruction error of signal) are trained. The main advantage of using analytical dictionary is to achieve the sparse coefficient with lower computational complexity. Result of first method by using validation on Colorado databaxse are 90%, 83.5%, 76.5% and 70% to classify two, three, four and five classes, respectively. The average results on Berlin databaxse by using validation is 97.56%. Result of second proposed method on Berlin test databaxse is 80.98%.