We present a computational model of the influence of lexico-syntactic gender on spoken-word recognition, and demonstrate its ability to account for relevant findings obtained with eye tracking (Dahan et al., 2000). The model is an SRN (Elman, 1990) trained on article-noun phrases input phoneme-by-phoneme. It learns to incrementally map its input to object concepts beginning with those sounds. After training, it exhibits behavior similar to French natives using gender to constrain lexical access: When the article preceding a noun is ambiguous in gender, all possible nouns are considered during lexical competition, but when a noun is preceded by a gender-marked article, only congruent nouns are considered as potential lexical candidates. The model is shown to generalize well to novel data including unseen article-noun combinations, leading us to conclude that it has in fact learned an abstract notion of gender and discovered the broader gender patterns in French article-noun sequences.