Experience with a second language (L2) has been shown to facilitate learning of a third or subsequent language (L3) (Sanz 2000). However, little is known about how much L2 experience is needed before benefits for L3 development emerge, or about whether effects depend on type of L3 instruction. We report two experiments investigating initial learning of semantic role (agent/patient) assignment in L3 Latin by native English speakers at different levels of formal experience in L2 Spanish (Beginning, Intermediate, Advanced, Very Advanced). In Experiment 1, (N = 76), learners engaged in a computerized Latin practice session that included metalinguistic information in feedback. In Experiment 2, (N = 0), feedback indicated only if learners' responses in practice were right or wrong. Development was operationalized as improvement in accuracy and/or reaction time. Results showed that when feedback was metalinguistic, an intermediate level of formal L2 experience was enough to yield an advantage, however only those learners with very advanced L2 experience showed longer-term retention of that advantage. In contrast, in the absence of metalinguistic feedback, very advanced experience was required for any advantage to appear. These results inform classroom accommodations for learners with different levels of previous language experience.