Judgments of others' emotional states and inclinations had recurrent fitness consequences for ancestral humans. Such judgments, made under uncertainty, can result in false positives (overestimating an emotional state or trait) or false negatives (underestimating the same). When the costs of these two errors consistently and historically differed, natural selection will have favored a bias toward making the less costly error. The perception of anger entails such asymmetry. Averaged across situations over evolutionary time, underestimation of anger was the more costly error, as the fitness decrements resulting from physical harm or death due to insufficient vigilance were greater than those resulting from lost social opportunities due to excessive caution. I therefore hypothesized that selection has favored an overestimation bias in the evaluations of others' state anger and general anger-proneness (trait anger) relative to evaluations of other traits to which this error asymmetry does not apply. Moreover, I hypothesized that additional attributes which make the actor more dangerous or make the observer more vulnerable increase the error asymmetry with regard to inferring state anger and trait anger, and should therefore correspondingly increase this overestimation bias. In the study described in Chapter 2 (N = 292), models photographed with a neutral expression and holding a potentially dangerous household object were judged to be angrier than those holding a harmless object, even though the held object provided no normative information about their emotional state. In Chapter 3, Study 1 (N = 161), a fictitious individual depicted in a vignette was judged to have higher trait anger than trait disgust even after controlling for the raters' perception of how emotionally he behaved. Moreover, trait anger ratings were more responsive than trait disgust ratings to behavioral cues of emotionality. In Chapter 3, Study 2 (N = 335), participants viewed images of angry or fearful faces. The interaction of factors indicating target's formidability (male sex), target's intent to harm (direct gaze), and perceiver's vulnerability (being female or high belief in a dangerous world) increased ratings of the target's trait anger but not trait fear. In aggregate, these results suggest a domain-specific bias in evaluating anger.