Corrective feedback (CF) can be provided to learners in different ways (explicit or implicit, focused or unfocused) and is the subject of major controversies in second language acquisition research. As no clear consensus has been reached so far about the most effective approach to CF with a view to fostering accuracy in second language (L2) writing -- and as this also holds true for blended learning environments -- our study aims at investigating the optimal conditions for effective CF in an experimental blended learning English as a Foreign Language (EFL) course. To explore this question, 93 study participants from Sorbonne Université (Paris, France) were divided up into seven groups: six treatment groups which received six different online CF types and one control group which received no CF. A performance comparison in terms of accuracy between the study participants' first and last pieces of writing was carried out to analyse the relative effectiveness of the different CF strategies. The results show that any type of CF is better than no CF at all, and that the repeated provision of unfocused indirect CF (with metalinguistic comments on the nature of errors) combined with extra computer-mediated micro-tasks over a certain period of time (24 weeks) seems to be the most efficient CF type in our context. This also points to the usefulness of integrating tutorial CALL activities in the CF provided as these seem to impact output accuracy positively.