This study was conducted in an authentic university setting with fourth-year Educational Sciences' students operating as online peer tutors to facilitate freshman tutees' online collaboration and knowledge construction in a blended "Instructional Sciences" course. Taking into account prior research uncovering weaknesses in online peer tutor behaviour, the aim of the study was to explore the possibility to optimise the quality and the nature of online peer support. In this respect, the study examined the impact of three tutor training conditions (multidimensional support, model/coach, and control condition) on peer tutors' actual tutor behaviour in asynchronous discussion groups and on tutors' self-efficacy beliefs, perceived collective efficacy, and training evaluation. Quantitative content analysis was applied to study online peer tutor behaviour. More specifically, two coding schemes were used focusing respectively on the occurrence of different e-moderating activities and on the evolution from modeling to coaching behaviour. Tutors' self-efficacy beliefs, perceived collective efficacy, and training evaluation were assessed by means of questionnaires. The results indicated that, compared to the control condition, in both the multidimensional support and the model/coach training the occurrence of social postings decreased whereas the presence of support stimulating tutees' personal development increased. Regarding the evolution from modeling to coaching behaviour, tutors started as a model in both experimental training conditions. Further, it appears that the experimental training conditions differ significantly with regard to peer tutors' self-efficacy beliefs on fostering knowledge construction and with regard to the tutors' personal training evaluation. Overall, it can be concluded that an explicit tutor training appears to determine the adoption of the expected types of tutoring activities. In this respect, providing novice peer tutors with guidelines by means of a specific training is fruitful for realising more adequate online tutoring behaviour and optimising self-efficacy beliefs regarding tutoring competences. (Contains 4 tables and 2 figures.)