Neuropsychological models postulate that the memory representation acquired for use in reading words is separate from the one acquired for use in spelling, while developmental models assume that the same representation is developed for access in both reading and spelling. The dual-representation model contends that there is often more precise information in reading representations than in spelling representations. This claim was tested in the current study using adult native speakers of English. People were supplied with minimal visual feedback while they spelled words that they knew and could read, and were then shown their whole spelling and asked whether they could improve upon it. People detected spelling mistakes on fewer than one in six trials after the reading check. They also returned many spellings to the original form, and were unable to improve upon them any more often than to change them to something equally bad or worse. The findings favour the view that normal individuals acquire a single orthographic representation from repeated exposures to a word during both reading and spelling. The representation may be adequate to permit successful reading but be insufficient for reproduction of the word-specific knowledge required for accurate spelling.