Research in BMIs has grown rapidly in the last few years. However, little attention has been paid to the overall system behavior, most published work being focused on the signal classification (i.e., translation) stage. More recently an increasing amount of work has been centred around the feature selection stage that precedes translation. The emphasis in feature selection and translation has stemmed from the large number of researchers with a machine learning or pattern recognition background who have recently joined the field. While there is an important contribution to BMIs, two crucial elements have not been sufficiently explored: the selection of suitable mental tasks and feedback protocols. This review presents an overview of BMIs and its main elements, with a focus on why each stage is important for the overall performance of such systems.