This exploratory study examined the relationship between corrective feedback (CF) and linguistic target complexity. In a pre-test/post-test/delayed post-test design, 44 adult intermediate L2 Italian learners from different L1 backgrounds were assigned to a didactic recast, a prompt and a no-feedback group. They were compared on oral and written measures on the development of passato prossimo, an Italian compound past form characterised by a set of complex semantic and morphosyntactic rules and participles displaying different degrees of form-meaning transparency. Mixed-effects models elucidated the extent to which feedback frequency predicted accuracy, whilst controlling for the effect of individual difference covariates and random variation. Only the frequency of didactic recasts predicted development of full passato prossimo sentences, whereas both feedback types were significantly related to participle development, a single aspect of the construction. Furthermore, only prompt frequency was positively related to accuracy in participles displaying more transparent (less complex) form-meaning relationships.