Inferring others' mental states and emotions is central to social cognition, yet research in this domain has predominantly focused on external cues such as facial expressions and body posture. Here, we introduce the Bodily Maps of Mental States Test (BMMST), a novel instrument assessing the ability to infer mental states from holistic representations of bodily sensations. In Study 1 (n = 250), participants generated Bodily Sensation Maps (BSMs) by indicating regions of increased and decreased bodily activity for 65 mental states, revealing spatially differentiated patterns and significant inter-rater reliability for a subset of states. In Study 2 (n = 286), an independent sample inferred mental states from these maps in a five-alternative forced-choice task. Mean accuracy reached 44%, well above chance (20%), with top performers exceeding 60%. Inference accuracy varied across states and was higher for those with more distinctive and reliable somatic patterns. Errors were systematically influenced by both perceptual similarity between maps and semantic relatedness between states, each contributing unique variance. Convergent validity was supported by associations with individual differences: performance was negatively related to externally oriented thinking (alexithymia), and accurate inference of "anxiety" was associated with higher scores on the corresponding personality facet. These findings show that mental states are grounded in shared embodied representations that can be decoded by others, and introduce the BMMST as a novel tool to investigate these mechanisms in social cognition, with potential clinical applications.