Two studies examined the notion that negative affectivity (Watson & Clark, 1984) is associated with more accurate perceptions of conveyed impressions in social interactions. In Study 1 (n = 160), low self-esteem (LSE) and high self-esteem (HSE) subjects were paired with either an LSE or an HSE partner. After a 15-min interaction, they rated themselves, their partners, and how they believed their partners would rate them on 20 adjectives related to social competence. Study 2 (n = 40) was identical except that each interaction was observed by 2 observers who rated each participant, and participants also rated how they believed an observer would rate them. LSE subjects exhibited greater accuracy only with respect to the elevation component of observers' ratings; HSE subjects overestimated the positivity of observers' evaluations, whereas LSE subjects were relatively accurate. However, LSE subjects exhibited less overall accuracy with respect to their partners' ratings. We argue that when these results are considered with earlier research, there is no support for the notion of depressive realism in assessing conveyed impressions.