We examined processing of written irony by recording readers' eye movements while they read target phrases embedded either in ironic or non-ironic story context. After reading each story, participants responded to a text memory question and an inference question tapping into the understanding of the meaning of the target phrase. The results of Experiment 1 (N = 52) showed that readers were more likely to reread ironic than non-ironic target sentences during first-pass reading as well as during later look-backs. Experiment 2 (N = 60) examined individual differences related to working memory capacity (WMC), Sarcasm Self-Report Scale (SSS), and need for cognition (NFC) in the processing of irony. The results of Experiment 2 suggest that WMC, but not SSS or NFC, plays a role in how readers resolve the meaning of ironic utterances. High WMC was related to increased probability of initiating first-pass rereadings in ironic compared with literal sentences. The results of these two experiments suggest that the processing of (unconventional) irony does require extra processing effort and that the effects are localized in the ironic utterances.