Abstarct: Several diseases can cause damage to the neuromuscular system, through which the brain is able to communicate and control the external environment. Recent advances in hardware and signal processing, using EEG signals is possible to the communication between human and computer. brain-computer interface (BCI), is a communication system to transmit messages from one person to the external world is through the normal exit routes brain, the peripheral nerves and muscles would be taken. One of the most popular feature extraction algorithms for brain-computer interface (BCI) baxsed on electroencephalogram (EEG), is using a common spatial pattern (CSP). Despite the popularity, efficiency and widespread use of CSP for BCI design, CSP is also known to be highly sensitive to noise and to severely overfit with small training sets. Hence, regular and improved methods of common spatial pattern as a way to extract the desired feature is more accurate. The methods presented in this study we have as an extension that Applicable on data from four classes and two classes of limitations that exist in many articles wiped out. Another algorithm is presented in this paper is divided into given time. For this purpose, data is divided into intervals and then filtered on any time frxame imposed CSP and its features are extracted. In this case, the effect of noise is much less We used dataset 2a of BCI competition IV for evaluate our method. This data set consists 22 channel of EEG signals from 9 subjects were recorded and sampled with 250 Hz. Finally we compare 4 RCSP algorithms to data sets of BCI competition. The results shows that Tikhonov regularized common spatial pattern (TRCSP) has better performance than the other methods.