This article addresses a current theoretical debate between modular and interactive accounts of sarcasm processing, by investigating the role of context (specifically, knowing that a character has been sarcastic before) in the comprehension of a sarcastic remark. An eye-tracking experiment was conducted in which participants were asked to read texts that introduced a character as being either sarcastic or not and ended in either a literal or an unfamiliar sarcastic remark. The results indicated that when the character was previously literal, a subsequent sarcastic remark was more difficult to process than its literal counterpart. However, when the context was supportive of the sarcastic interpretation (i.e., the character was known to be sarcastic), subsequent sarcastic remarks were as easy to read as literal equivalents, which would support the predictions of interactive accounts. Importantly, this effect was not preceded by a main effect of literality, which constitutes evidence against the predictions of modular accounts.