Brain-computer interfaces (BCIs) allow persons with impaired mobility to communicate and interact with the environment, supporting goal-directed thinking and cognitive function. Ideally, a BCI should be able to recognize a user's internal state and adapt to it in real-time, to improve interaction. Our aim was to examine studies investigating the recognition of affective states from neurophysiological signals, evaluating how current achievements can be applied to improve BCIs. Following the PRISMA guidelines, we performed a literature search using PubMed and ProQuest databases. We considered peer-reviewed research articles in English, focusing on the recognition of emotions from neurophysiological signals in view of enhancing BCI use. Of the 526 identified records, 30 articles comprising 32 studies were eligible for review. Their analysis shows that the affective BCI field is developing, with a variety of combinations of neuroimaging techniques, selected neurophysiological features, and classification algorithms currently being tested. Nevertheless, there is a gap between laboratory experiments and their translation to everyday situations. BCI developers should focus on testing emotion classification with patients in ecological settings and in real-time, with more precise definitions of what they are investigating, and communicating results in a standardized way.