Some patients with pure alexia or letter-by-letter reading demonstrate the Saffran effect: residual activation of higher order lexical-semantic representations despite poor word recognition. This study investigated the reading of patient FD, a letter-by-letter reader with a clear Saffran effect. Two alternative explanations for this effect were tested in a series of experiments and through the impact of whole-word and letter-based therapies on FD's reading. One theory assumes that the disparity between overt recognition and implicit activation of word meaning is underpinned by two separate reading systems. An alternative hypothesis argues for a single whole-word reading system supplemented by the deliberate, compensatory strategy of letter-by-letter reading. Under this hypothesis, the Saffran effect reflects partial activation of the single, whole-word system. FD's results strongly supported the latter hypothesis. FD's reading behaviour was characterised by partial activation of higher word representations, accuracy was graded by word variables known to influence the normal reading system, and most importantly, once the characteristics of the tasks were equated, there was no evidence for a dissociation between word categorisation and recognition. In addition, the whole-word therapy encouraged FD to abandon the letter-by-letter strategy. Without this compensatory technique, FD's emergent deep dyslexia was consistent with a partially activated, whole-word reading system that produces overt reading responses. Comparison of data from this and other studies suggests that the Saffran effect is most likely to be observed in patients with severe pure alexia. © 2004 Psychology Press Ltd.