READING and writing require access to stored knowledge about the spelling of words. Presumably, we recognize chair but not chare or chiar as a word of English, and similarly would write 'chair' but not 'chare' or 'chiar', because we access orthographic representations that specify the identity and the order of the graphemes (abstract letter representations) that comprise the spelling of words1,2. Thus, a fundamental problem concerns the content and structure of the hypothesized orthographic representations, and how information about grapheme order is represented and processed. We present evidence from a brain-damaged patient (N.G.) with unilateral neglect that this information is coded spatially. Unilateral neglect is a disorder clinically characterized by the inability to perceive or respond to stimuli presented to the side contralateral to the site of lesion, despite the absence of significant sensory or motor deficits3,4. The patient made reading and spelling errors only on the right half of words, regardless of length. Furthermore, she produced the same pattern of errors in reading and spelling, irrespective of the topographic arrangement of stimuli in reading (horizontal, vertical or mirror-reversed words) and of the type of response in spelling (written, oral or backward oral spelling). This pattern of performance suggests that order information in orthographic representations is coded spatially in a word-centred coordinate system; that is, in a spatially defined coordinate frame whose centre corresponds to the midpoint of a canonical, orientation-invariant representation of the word and not the midpoint of the word stimulus5.