A comparison of the error profiles of monolingual (child L1) learners of Dutch, Moroccan children (child L2) and Moroccan adults (adult L2) learning Dutch as their L2 shows that participants in all groups massively overgeneralize [-neuter] articles to [+neuter] contexts. In all groups, the reverse gender mistake infrequently occurs. Gender expressed by Dutch attributive adjectives reveals an age-related asymmetry between the three groups, however. Whereas participants in the child groups overgeneralize one particular suffix (namely the schwa), adult participants use both adjectival forms, the schwa-adjective and the bare adjective, incorrectly. It is argued that the asymmetry observed in adjectives reflects that adult learners exploit an input-based, lexical learning route, whereas children rely on grammar-based representations. The similarity in article selection between all groups follows from the assumption that adults, like children, make use of lexical frames. Crucially, lexical frames can successfully describe the distribution of gender-marked articles, but they cannot account for gender in adjectives. Appended are: (1) Information about participants is given in Tables 6 and 7; and (2) Consistency analysis. (Contains 10 tables and 2 figures.)