In two experiments, we examined if single emotion recognition is improved by a training to assign bodily maps (i.e., topographically distinct and culturally universal bodily sensations) to corresponding emotions. In Experiment 1, participants in the experimental group (N = 38) viewed emotion-arousing pictures eliciting predominantly one specific emotion (i.e., fear, anger, disgust, and sadness) and learned to assign emotions to emotion-specific bodily sensations. Control group participants (N = 36) were exposed to the same set of pictures and judged the content of the pictures on semantic dimensions (e.g., animals, people). Subsequently, a set of emotion-arousing pictures was presented, and all participants were asked to estimate the intensity of each emotion. Training of emotion-specific bodily maps improved single emotion recognition relative to the control group. Experiment 2 (N = 43) replicated the findings of Experiment 1 and suggests that the training effect is caused by semantic processes.