A recent focus in the development of music recommendation systems is the incorporation of physiological signals. Among this, the possibility of using non-invasive, electroencephalography-based neural activity is of great interest. In this preliminary work, we sought to predict the preference of individuals for previously unheard music through a combination of acoustic and neural features. We developed a real-time system for preference decoding which was used to skip songs with ∼80 ms latency according to users’ desires. The results suggest that music recommendation systems could supplement acoustic features with neural activity for characterising an individual’s music preferences in real time, with options to incorporate further acoustic and physiological information for improved