Language relies on the rule-based combination of words with different grammatical properties, such as nouns and verbs. Yet most research on the problem of word retrieval has focused on the production of concrete nouns, leaving open a crucial question: how is knowledge about different grammatical categories represented in the brain, and what components of the language production system make use of it? Drawing on evidence from neuropsychology, electrophysiology and neuroimaging, we argue that information about a word's grammatical category might be represented independently of its meaning at the levels of word form and morphological computation.Understanding how words are represented and retrieved in the brain will be crucial to any eventual neurobiological theory of language. The problem goes beyond determining how neural circuits encoding a particular conceptual representation are mapped to a unique and arbitrary set of representations for sounds, letters, or gestures: although this might suffice for the production of isolated words, it fails to specify how words are combined into phrases and sentences. How do we know, for example, that 'the singing are birds' is not a well-formed alternative to 'the birds are singing'? (And why are such exchanges rare even among speech errors? [1]) The most obvious answer is that bird and sing belong to different grammatical categories -noun and verb, respectively -with different syntactic roles, and that this information is encoded in the brain and accessed during the course of language production.Even though grammatical categories are not created equal [2], most research in the area of lexical access has used the production of concrete nouns as a proxy for word production in general. Therefore, it is not too surprising that our current picture of the cortical organization of lexical knowledge leaves little room for mechanisms that might be sensitive to differences between words of different categories. A recent review of neuroimaging studies of lexical access parcels the left perisylvian cortex into areas involved in retrieving word forms from concepts and in various stages of phonological processing [3], but sheds little light on the question of how and where the brain processes a word's grammatical role.This picture might change as more researchers use neuroscientific methods to address finer-grained questions about the representation of words in the brain. To date, however, experiments on the retrieval of grammatical category information have yielded conflicting results. Although electrophysiological measures have suggested that nouns and verbs are processed by distinct neural generators, neuroimaging has turned up little evidence for an anatomical distinction. The neuroimaging results have led some to argue that words are stored in distributed networks, in which categorical distinctions are not explicitly represented [4] -a conclusion that is at odds with what is known from the study of patients with language deficits following brain damage. Are we at an empirical impasse...