Subjects read scenarios where a speaker made a comment that, depending on information that was privileged to the subjects, could have been interpreted as sarcastic or not sarcastic. Their task was to take the perspective of an uninformed addressee and determine whether he or she would perceive sarcasm. Overall, when subjects believed that the speaker was actually being sarcastic they were more likely to attribute the perception of sarcasm to the addressee--even when the message was conveyed in writing and could not have involved disambiguating cues such as intonation. Data from four different experiments suggest that readers do use information that is perspective-irrelevant and thus pose a problem for theories of language use that assume readers only use "relevant" information.