Comprehension of unfamiliar metaphors (mercy blanket) is an effortful cognitive process that requires the formation of a novel metaphoric mapping between two disparate domains during which irrelevant properties have to be suppressed. The present study aims to examine the relationship between the comprehension of both unfamiliar and familiar metaphors and working memory. Three experiments were conducted: a comprehension task (Experiment 1), a recognition task (Experiments 2a and 2b), and a free recall task (Experiment 3). In the first experiment comprehension of both unfamiliar and familiar metaphors correlated with digit span backward but not with digit span forward. Results of the second experiment revealed that unfamiliar metaphors induced a higher rate of semantic errors relative to phonological errors, whereas familiar metaphors induced the same number of phonological and semantic errors. The third experiment confirmed that unfamiliar metaphors were harder to recall than were familiar metaphors. These findings show that working memory capacity may be involved in the computation of unfamiliar metaphoric interpretations, and more specifically in the process of suppressing irrelevant information via the central executive. Familiar metaphor recognition may rely on either phonological codes that are maintained in the phonological loop or on semantic processing that involves long term storage.