According to recent models of word production, when we name a picture, we first retrieve the meaning of the object, and then we independently retrieve the written or sound form of the word corresponding to the picture. In languages like French, in which words have a gender, theoretical models disagree with respect to the moment at which this information is retrieved. The lemma model (Levelt et al., 1999) posits that we access this information before the sound or written form of the word is retrieved. In contrast, the"Independent network"(IN) model (Caramazza, 1997) model posits that we access gender after retrieval of either the sound or written form of the word. This paper reports a single-case study of an aphasic patient, BA, who showed deficits affecting spoken and written production in the presence of largely preserved comprehension abilities. Experimental testing indicated that she presented with a deficit functionally localized in the access to lexical representations. Results in picture naming and in gender identification also revealed that BA identified the gender above chance level, whether she produced a correct response, a phonemic error, or a neologism. In contrast, when she was unable to produce a spoken or written response, she could not identify the gender. This pattern of performance is consistent with the lemma model in which access to lexical syntax is required before access to phonological form can take place.