Current models of word production provide different accounts of the representations of homophones--words that sound the same but have different meanings (e.g., muscle/mussel; (a) walk/(to) walk). A point of disagreement concerns frequency: While some models assume that homophone processing varies as a function of the frequency of the individual homophonic forms, other models predict that the combined frequency of the homophonic forms (e.g., the frequency of muscle+mussel) determines how homophones are processed. These contrasting views were tested in a series of experiments with AW, an English-speaking brain-damaged woman who showed anomia, a deficit of word phonology retrieval in speech production. AW's semantic processing was intact. In oral naming, we observed a frequency effect: AW was significantly more successful in producing high- as opposed to low-frequency words. Our results consistently demonstrated that AW's successful naming reflected the frequency of the individual homophonic forms, rather than the combined frequency of the homophonic forms. Our results provide support for models of speech production that identify the frequency of the individual homophones as the critical factor in homophone naming.