Computerized cognitive training has the potential to be an important tool for both rehabilitation and prevention of cognitive impairment. However, the demonstrated effects of cognitive training have often been limited, and seemingly conflicting results have been reported. This paper reviews selected results from several disciplines that relate to the challenges and the potential of computerized cognitive training. Developments within cognitive neuroscience and human-computer interaction (HCI) are related to virtual rehabilitation, and to a need to develop training applications that are both adaptive and realistic. Efficient rehabilitation and training depends on an ability to monitor the development of targeted functions. Monitoring cognitive functions is particularly challenging because of limited access to, and understanding of, these functions. Psychophysiological computing and brain-computer interfaces (BCIs) are reviewed as promising approaches for monitoring cognitive functions. Interpretation and understanding of such measurements should be supported by a basis in cognitive neuroscience and theories of brain function. This paper focuses on some general themes in such theories, with a strong connection to the importance of realism and interaction with reality. Such theoretical results are related back to practical applications through brain imaging studies concerning realistic interaction. To summarize the interdisciplinary material reviewed here the concept of reality-based brain-computer interaction (RBBCI) is introduced. RBBCI is intended to capture important principles for the development of realistic and adaptive systems for cognitive rehabilitation, and to serve as a cornerstone concept for interdisciplinary development.