Studies on cross-linguistic syntactic priming suggest that bilinguals can share syntactic representations across languages (e.g., Hartsuiker, Pickering, & Veltkamp, 2004). But how are these representations established in late learners of a second language? Specifically, are representations of syntactic structures in a second language (L2) immediately collapsed with similar structures of the first language (L1), or are they initially represented separately? In order to investigate this, we primed the use of English genitives with Dutch (Experiment 1) and English (Experiment 2) genitives (e.g., "het hemd van de jongen/the shirt of the boy" vs. "de jongen zijn hemd/the boy's shirt") in late Dutch-English bilinguals with varying levels of proficiency in English (their L2). The head nouns of prime and target constructions either had the same meaning ("hemd/shirt--shirt") or a different meaning "(duim/thumb--shirt"), in order to test whether the use of both genitives was generalized across nouns. Experiment 1 found stronger between-language priming for more than less proficient bilinguals in both conditions, thus suggesting a shift from language-specific to shared syntactic representations. Experiment 2 suggests that these early, language-specific syntactic representations might be item-specific: Less proficient bilinguals showed much weaker priming when the heads of prime and target constructions had different meanings than when they were repeated. (Contains 6 tables and 8 figures.)