Priming studies investigating the processing of emotional faces under conditions of limited awareness have shown that people can extract more than just valence from masked faces. However, previous results have been inconsistent with regard to the degree of differentiation among negative expressions. Some results have suggested a relevance differentiation (i.e., anger differentiated from fear or sadness) and some have suggested differentiation by arousal (i.e., sadness differentiated from fear or anger); others have not suggested any differentiation beyond valence. It may even be possible that differentiation occurs down to the level of the specific emotion. To gain further insight into emotion differentiation under such conditions, we presented angry, fearful, and sad faces as masked primes in a response priming task with nontarget primes (i.e., primes and targets were from different stimulus sets). More important, in each of four experiments, only 2 of the prime emotions were used as target emotions in a binary emotion categorization task; that is, 1 prime emotion was left out as a target category. The relevance, arousal, and specificity hypotheses make contrasting predictions regarding (a) the presence or absence of focal priming effects (i.e., effects from prime emotions contained in the response set) and (b) the presence or absence of priming effects arising from the prime emotion left out from the response set. Results of conventional analysis of response times as well as diffusion model analyses were most compatible with the specificity hypothesis. However, the particular response set partially determined which information was extracted from masked primes. Results are interpreted in terms of an action-trigger account. (PsycINFO Database Record (c) 2018 APA, all rights reserved).