The claim that certain facial expressions signal certain specific emotions has been supported by high observer agreement in labeling the emotion predicted for that expression. Our hypothesis was that, with a method common to the field, high observer agreement can be achieved through a process of elimination: As participants move from trial to trial and they encounter a type of expression not previously encountered in the experiment, they tend to eliminate labels they have already associated with expressions seen on previous trials; they then select among labels not previously used. Seven experiments (total N = 1,068) here showed that the amount of agreement can be altered through a process of elimination. One facial expression not previously theorized to signal any emotion was consensually labeled as disgusted (76%), annoyed (85%), playful (89%), and mischievous (96%). Three quite different facial expressions were labeled nonplussed (82%, 93%, and 82%). A prototypical sad expression was labeled disgusted (55%), and a prototypical fear expression was labeled surprised (55%). A facial expression was labeled with a made-up word (tolen; 53%). Similar results were obtained both in a context focused on demonstrating a process of elimination and in one similar to a commonly used method, with 4 target expressions embedded with other expressions in 24 randomly ordered trials.