Abstract Using a database of vibratory signals captured from the index finger of participants performing self-touch or touching another person, we wondered whether these signals contained information that enabled the automatic classification into categories of self-touch and other-touch. The database included signals where the tactile pressure was varied systematically, where the sliding speed was varied systematically, and also where the touching posture were varied systematically. We found that using standard sound feature-extraction, a random forest classifier was able to predict with an accuracy greater than 90% that a signal came from self-touch or from other-touch regardless of the variation of the other factors. This result demonstrates that tactile signals produced during active touch contain latent cues that could play a role in the distinction between touching and being touched and which could have important applications in the creation of artificial worlds, in the study of social interactions, of sensory deficits, or cognitive conditions.