Humans consistently face the challenge of discerning liars from truth-tellers. Hundreds of studies in which observers judge the veracity of laboratory-created lies and truths suggest that this is a difficult task; in this context, lie-detection accuracy is notoriously poor. Challenging these findings and traditional methodologies in lie-detection research, we draw upon the somatic marker hypothesis and research on interoception to find that: (a) people experience physiological reactions indicating increased sympathetic arousal while observing real, high-stakes lies (vs. truths), and (b) attending to these physiological reactions may improve lie-detection accuracy. Consistent with the tipping point framework, participants demonstrated more physiological arousal and vasoconstriction while observing real crime liars versus truth-tellers, but not mock crime liars versus truth-tellers (Experiment 1; N = 48). Experiment 2 replicated this effect in a larger sample of participants (N = 169). Experiment 3 generalized this effect to a novel set of stimuli; participants demonstrated more physiological arousal to game show contestants who lied (vs. told the truth) about their intention to cooperate in a high-stakes economic game (N = 71). In an intervention study (Experiment 4; N = 428), participants were trained to attend to their physiological signals; lie-detection accuracy increased relative to a control condition. Experiment 5 (N = 354) replicated this effect, and the addition of a bogus training condition suggested that increased accuracy was not simply attributable to self-focused attention. Findings highlight the limitations of relying on laboratory-created lies to study human lie-detection and suggest that observers have automatic, physiological reactions to being deceived.