A fundamental question in language acquisition research is whether young children have abstract grammatical representations. We tested this question experimentally. French-learning 30-month-olds were first taught novel word-object pairs in the context of a gender-marked determiner (e.g., un[subscript MASC]ravole "a ravole"). Test trials presented the objects side-by-side while one of them was named in new phrases containing other determiners and an adjective (e.g., le[subscript MASC]joli ravole[subscript MASC] "the pretty ravole"). The gender agreement between the new determiner and the nonadjacent noun was manipulated in different test trials (e.g., le[subscript MASC]__ravole[subscript MASC]; *la[subscript FEM]__ravole[subscript MASC]). We found that online comprehension of the named target was facilitated in gender-matched trials but impeded in gender-mismatched trials. That is, children assigned the determiner genders to the novel nouns during word learning. They then processed the non-adjacent gender agreement between the two categories (Det, Noun) during test. The results demonstrate abstract featural representation and grammatical productivity in young children.