This study describes a therapy programme that aimed to improve oral reading and spoken naming in a deep dyslexic patient by teaching the patient grapheme-phoneme correspondences. This aim was achieved and the improvement could be attributed to the specific effects of the therapy. However, the pattern of results obtained was very different to that predicted. The patient proved unable to perform or to relearn one of the processes necessary for successful reading by a sublexical routine (blending of individual phonemes to form a word). It is concluded that the improvement had been achieved not by a learned mechanism for sublexical reading but instead by enabling the use of patient-generated phonemic cues. The therapy had enabled the patient to produce successfully the phoneme associated with the initial letter of the word, which then acted as a phonemic cue facilitating reading of that word. Spoken naming was also facilitated as information on the written form of the word could similarly be used to provide a phonemic cue for spoken production.