Background: The characteristics of graphemic buffer disorder have been described by Miceli, Silveri, and Caramazza (1985) and Caramazza and Miceli (1990) and include a length effect in spelling words and nonwords, in both written and oral format. Error patterns typically consist of omissions, substitutions, additions, and movement errors. Recently, lexical effects on spelling accuracy in many buffer cases have been shown (Sage & Ellis, 2004) and, in patient BH in particular, these included an influence of orthographic neighbourhood size (the number of words that can be generated by changing one letter in the target word; Coltheart, Davelaar, Jonassen, & Besner, 1977). Aims: This paper aims to show that orthographic neighbours used in a therapy programme can bring about improvements in the spelling of targets that have not directly received therapy. Methods & procedures: This was a single case treatment study of BH, who showed classic features of graphemic buffer disorder in her spelling. Two priming studies contrasting no prime, control primes, and orthographic neighbour primes (both word and nonword) established that positive effects on spelling accuracy and error pattern could be achieved. The use of orthographic neighbours was then extended into a therapy programme. Three word sets contrasted direct therapy to the words themselves (Set1), no therapy at all (Set 2), and therapy to neighbours of the words in a set (Set 3). An errorless learning paradigm was used throughout the therapy programme. Outcome & results: Improvement was made both to the treated words in Set 1 and to the words in Set 3, even though these words had received no direct therapy. There was no change in accuracy in Set 2, the control set that had received no therapy. The paper also explores changes in error patterns due to therapy, showing that error patterns changed following therapy. Conclusions: The priming studies showed that changes in accuracy could be achieved when orthographic neighbours were used. Following on from this, a therapy programme based on neighbours was effective in assisting the graphemic buffer. Specifically, the interaction between lexicon and buffer was used therapeutically to improve not only the spelling of words that had received direct treatment but also to a word set that did not directly receive treatment. These changes were brought about using an errorless learning paradigm.