College students wrote definitions of either abstract or concrete nouns in longhand while performing a concurrent working memory (WM) task. They detected either a verbal (syllable), visual (shape), or spatial (location) stimulus and decided whether it matched the last one presented 15-45s earlier. Writing definitions of both noun types elevated the response time to verbal targets above baseline. Such interference was observed for visual targets only when defining concrete nouns and was eliminated entirely with spatial targets. The interference effect for verbal targets was the same whether they were read or heard, implicating phonological storage. The findings suggest that language production requires phonological or verbal WM. Visual WM is selectively engaged when imaging the referents of concrete nouns.