Background: Identifying the point of breakdown in people with aphasia with disorders of word retrieval is not straightforward. Evidence has been sought from: (i) the nature of the errors in naming; (ii) the variables affecting naming accuracy; (iii) the effects of correct and misleading cues; (iv) performance in other word comprehension and production tasks. However, previous research has demonstrated that each of these sources of evidence provides information compatible with more than level of breakdown. Aims: The study investigates whether a combination of information from these sources can provide a coherent account of how word retrieval is breaking down in people with aphasia. Methods & Procedures: Three people with aphasia (JGr, LM, and KS) took part in four experiments. The first investigated the errors made in picture naming and the factors (target word length, imageability, frequency...) affecting naming accuracy. The second experiment investigated the effects of correct phonemic cues and miscues on word retrieval. The third examined the participants' performance in tests of spoken and written word and picture comprehension. The fourth experiment investigated whether the participants had the processing abilities necessary to generate their own phonemic cues in spoken naming from orthographic information. Outcomes & Results: Evidence from these investigations showed different levels of breakdown in the three participants. JGr's naming was characterised by semantic errors, effects of target imageability and familiarity on naming accuracy, improved naming with correct phonemic cues and semantic errors with miscues, and poor performance in word comprehension tasks. This pattern is consistent with a breakdown at a semantic level underlying JGr's difficulty in word retrieval. In contrast, LM shows performance indicating a breakdown in mapping between intact semantic and phonological representations. He makes primarily no response errors in naming and his accuracy is affected only by frequency and familiarity. Correct phonemic cues can improve his naming accuracy to near normal levels, and he makes no semantic errors, although he is slowed by miscues. His word and picture comprehension is intact. KS shows a more complex pattern of impairment. Like JGr, she shows evidence of a semantic impairment: she makes semantic errors in naming, and her accuracy is affected by target imageability. She makes errors in word comprehension and her word retrieval is adversely affected by miscues. There are two unusual features to her performance: her naming accuracy is not improved by initial phoneme cues (despite effects of miscues and more extensive phonemic cues), and she is better at naming pictures with longer names (a "reverse length effect"). Investigations in experiment four show that KS is using orthographic information on the initial letter of names to generate her own phonemic cues; it is concluded that in addition to her semantic deficit she has an impairment in access to lexical phonological representations. Conclusions: We conclude that careful investigation of the performance of people with aphasia across a range of tasks can be used to identify underlying levels of breakdown in word retrieval. However, superficial resemblances between people with aphasia can be misleading.