In a cross-modal (auditory-visual) fragment priming study in French, we tested the hypothesis that gender information given by a gender-marked article (e.g. unmasculine or unefeminine) is used early in the recognition of the following word to discard gender-incongruent competitors. In four experiments, we compared lexical decision performances on targets primed by phonological information only (e.g. /kRa/-CRAPAUD /kRapo/; /to/-TOAD) or by phonological plus gender information given by a gender-marked article (e.g. unmasculine /kra/-CRAPAUD; a /to/-TOAD). In all experiments, we found a phonological priming effect that was not modulated by the presence of gender context, whether gender-marked articles were congruent (Experiments 1, 2, and 3) or incongruent (Experiment 4) with the target gender. Moreover, phonological facilitation was not modulated by the presence of gender context, whether gender-marked articles allowed exclusion of less frequent competitors (Experiment 1) or more frequent ones (Experiments 2 and 3). We concluded that gender information extracted from a preceding gender-marked determiner is not used early in the process of spoken word recognition and that it may be used in a later selection process.