Mapping brain activity patterns in external actions has been studied in recent decades and is the base of a brain-computer interface. This type of interface is extremely useful for people with disabilities, where one can control robotic systems that assist, or even replace, non functional body members. Part of the studies in this area focuses on noninvasive interfaces, in order to broaden the interface usage to a larger number of users without surgical risks. Thus, the purpose of this study is to assess the performance of different pattern recognition methods on the classification of mental activities present in electroencephalograph signals. Three different approaches were evaluated: Multi Layer Perceptron neural networks; an ensemble of adaptive neuro-fuzzy inference systems; and a hierarchical hybrid neuro-fuzzy model. © 2009 Springer Berlin Heidelberg.