Introduction Outcome-based approaches to education and the inherent emphasis on programmatic assessment in particular, require models of mentoring in which mentors fulfil dual roles: coach and assessor. Fulfilling multiple roles could result in role confusion or even role conflicts, both of which may affect mentoring processes and outcomes. In this study, we explored how mentors conceptualise and enact their role in a multiple-role mentoring system and to what extent they experience role conflicts. Methods We conducted a constructivist grounded theory study at one undergraduate medical school. A purposive sample of 12 physician-mentors active in a programmatic assessment system was interviewed. Data analysis followed stages of open, axial and selective coding through which themes were constructed. Results Three predominant mentoring approaches were constructed: (i) empowering (a reflective and holistic approach to student development); (ii) checking (an observant approach to check whether formal requirements are met), and (iii) directing (an authoritative approach to guide students' professional development). Each approach encompassed a corresponding type of mentor-mentee relationship: (i) partnership; (ii) instrumental, and (iii) faculty-centred. Furthermore, mentors' strategies, focus, agency provided to students and perception of the assessment system characterised mentoring approaches and relationships. Role conflicts were mainly experienced by mentors with a directing mentoring approach. They used various coping mechanisms, including deviation from assessment guidelines. Conclusions In multiple-role mentoring in the context of programmatic assessment, mentors adopted certain predominant mentoring approaches, which were characterised by different strategies for mentoring and resulted in different mentor-mentee relationships. Multiple-role mentoring does not necessarily result in role conflict. Mentors who do experience role conflict seem to favour the directing approach, which is most at odds with key principles of competency-based education and programmatic assessment. These findings build upon existing mentoring literature and offer practical suggestions for faculty development regarding approaches to mentoring in programmatic assessment systems.