Abstract: Music recommendation systems are researched from multiple aspects, and the recent focus has been on the relationship between physiological measures and music preference. These studies, however, face the challenge of varying acoustic features among stimuli, which can con-found with the effect of preference. Also, access to physiological signals is often limited in real-life applications such as smart devices. In this study, we aimed to reduce the effect of acoustic variability by presenting different expressions of the same musical piece while connecting the study to daily use by shortening the stimulus length. We predicted participants’ preference for musical expressions from cardiac and respiratory data measured from 30 subjects in a psychophysiological experiment. We identified a non-linear relationship between physiological signals and musical preference over an ultrashort time interval (~15 s). This result suggests that music recommendation systems can use biological signals to adapt their model rapidly on minimal data retrieval.