With rapid advances in digital technologies, the popularity of avatars - digital representations of people in computer-mediated environments - is growing. Avatars allow people to visually represent their offline social identity, or selectively render certain layer(s) of their social identity less identifiable or unidentifiable in online environments. The present research investigated how African Americans, whose racial identity often suffers negative stereotyping, responded to stereotype threat when they performed a stereotype-relevant task with 2 ostensible coactors in an avatar-represented group setting. A 2 x 2 between-participants design manipulated salience of racial identity (salient vs. nonsalient) and performance context (competition vs. cooperation), and assessed the extent to which participants persisted on an extremely challenging stereotyperelevant task (unsolvable anagram). The results showed that in the context of competition, participants in the race-nonsalient avatar group persisted significantly longer on the unsolvable anagram than did those in the racesalient avatar group; however, in the context of cooperation, no significant difference was found between the 2 avatar groups. The findings indicate that the effects of identity salience as varied by different types of avatars (identity-revealing vs. identity-cloaking) on identity-associated threat may be moderated by the contexts of performance in which the target individuals are situated.