Background: This single case study examines the linguistic phenomenon of ambiguous spoken words: homophones . In the psycholinguistic research literature the lexicalisation of homophones is the subject of extensive debate. A common assumption is that these words share one word form but have two grammatical representations (lemmas). An opposing view postulates two separate word form entries for homophones - without assuming a lemma level. Aims: The single case study presented here searches for empirical evidence for the representation of homophones using aphasic speech production. Can aphasic speech production give us some evidence regarding how many processing levels have to be completed prior to articulation? Methods & Procedures: A treatment study with MW, a man with global aphasia and severe anomia, is presented. Treatment comprised an intensive picture-naming training with exclusively phonological cues. Naming was facilitated using the following cueing hierarchy: (i) giving the initial phoneme, (ii) tapping the syllable number, and (iii) giving the target word for repetition. How this pure phonological training would affect naming performance of homophones, semantically and phonologically related words, and unrelated words was investigated. Outcomes & Results: The results showed significant short-term, item-specific effects for treated words and generalisation to untreated homophone words alone. The outcome is discussed with reference to the debate regarding homophone production in psycholinguistics and the debate regarding the facilitatory effects of phonological techniques. Conclusions: The results support the two stage model, with only one word form and two lemma entries for homophones. In addition, the outcome of this phonological treatment supports the common assumption that pure word form training rarely results in long-term improvement or generalisation.