A number of recent studies have attempted to explain patterns of normal and impaired performance in a variety of different language tasks with respect to the same set of "primary" systems rather than resorting to explanations in terms of dedicated processes, specific to each and every language activity. In this study we consider whether the same approach can be taken to patterns of impaired single-word speech production. Specifically, using cross-sectional data from 21 aphasic patients we tested the hypothesis that the degree and nature of anomia can be explained using independently derived measures of the integrity of the patients' phonological and semantic/conceptual representations, without postulating a role for an abstract lexical level of representation. At a global level, we found that these two measures explained 55-80% of the variance in the patients' naming accuracy, a figure which approaches that found for test reliability. There was also a close fit between observed and expected naming accuracy for all individual patients. The same two measures also predicted the rate of different types of anomic error across individuals. Measures intended to assess lexical integrity did not explain any additional, unique variance in naming accuracy. We discuss these results and the theoretical approach with respect to existing theories of speech production, and evaluate the case-series methodology itself, both as a tool to reveal the underpinnings of speech production and as a neuropsychological technique in general.