In four experiments (two in French and two in Italian), we investigated whether the language production system uses conceptual information regarding biological gender in the encoding of gender agreement between a subject and a predicate. Both French and Italian have a nominal gender system that includes a distinction between nouns reflecting the sex of the referent (conceptual gender) and nouns for which the gender does not reflect the sex of the referent (grammatical gender). The experiments used a constrained sentence completion task (Vigliocco, Butterworth, & Semenza, 1995). In Experiments 1 (Italian) and 2 (French) we found that errors in the agreement of gender between the subject and the predicate were more common when the subject head noun did not have any conceptual correlates. Experiments 3 and 4 established that the advantage for conceptual gender in the first two studies cannot be explained by the difference in animacy between nouns with conceptual gender (referring to humans and animals) and nouns with grammatical gender (referring to objects).