This master’s thesis is about the learning potential in mentoring conversations, which is investigated through an analysis of interpersonal communication. In Denmark, workplace mentoring is sometime used for onboarding of newly graduated social workers in municipalities, some experience a ‘practice shock’ (Jensen, 2015, p. 8). However, a literature review within the research field of workplace mentoring finds that hardly any studies are based on observa-tions of mentoring conversations. Therefore, this master’s thesis aims to contribute with insights on the learning potential in mentoring conversations. Motivated by the above, the primary data material of this master’s thesis consists of two video ob-served mentoring conversations: Each with a newly graduated and employed social worker as the mentee and with a more experienced social worker as mentor. Each mentoring pair are colleagues and work in the same team. The fact that both mentors have more professional expertise combined with the fact that both men-toring programs are to end after the first three months of the mentee’s employment, poses a theoreti-cal based curiosity: According to Schein (1999; 2010) a helper (is this case a mentor) can help (1) by using an expertise-based approach, which seem appropriate in the light of both mentors’ professional experience, but this approach can lead to dependability. Or the helper can help (2) by using a process consultant approach aiming to support the “capacity for learning so that it can in the future fix its own problems” (Schein, 1999, p. 19), but how then do the mentees benefit from their mentor’s expertise? With a focus on mentee’s learning potential in a mentoring conversation, this master’s thesis raises the question: “How can the mentor by virtue of her professional experiences, knowledge, and competen-cies, help the mentee with her challenges, and at the same time maintain mentee’s own responsibility for her own challenges?” (translated from paragraph 1.1.1). The primary analytical data material is supplemented with a semi-structured interview with each of the four observed social workers. To answer the problem statement, the hermeneutically inspired analysis process DiaLoop (Alrø, Dahl, and Schumann, 2016) is applied for analysis, as this master’s thesis’s phi-losophy of science is based on philosophical hermeneutics (Gadamer, 2007). The analysis is based on the theory of Schein (1999; 2010), theory on single-loop-, double-loop- and deutero-learning (Argyris 2003; 2012; Argyris & Schön, 1978), theory on the balancing of expectations (Alrø & Kristiansen, 1998), and theory of potential differences in perspectives in conversations (Alrø, 1996). Following an analysis and a discussion of the conversations, it is concluded that: An initial balancing of expectation is important, as neither the mentee’s present learning needs nor challenges – nor details hereabout – are given. The mentee may need help to learn a specific, profes-sional procedure, which would be achieved by single-loop-learning and an expertise-based approach. Or the mentee may need help to examine different possible actions, where the mentor could help by nurturing double-loop-learning and by using the process consultant approach. The process of balancing expectations should end with the mentee having ‘the last word’, as a means of letting the mentee be responsible for defining her own challenges. For the mentor to contribute with expertise and for the mentee to have the responsibility for her own challenges, it is important that the mentor can change between approaches during the conversation. A shift in approaches can be initiated by a disagreement on either part. Without such changes in the mentor’s approach it is impossible to collectively examine a disagreement constructively – which would be a hindrance to the learning potential. For the mentee to stay in charge of her challenges, the mentor needs to end a disagreement by using the process consultant approach. As the mentoring program is to end within three months, the mentor should focus om deutero-learning at both single- and double-loop-learning-level, for the mentee to learn how to handle future challeng-es. As metacommunication can strengthen deutero-learning, the mentor should be aware of how one is communicating. Furthermore, as a colleague the mentor can help the mentee to learn how things are done in their team, and hereby reduce potential learning transfer problems. But at the same time, the mentor should be aware of the ‘manualisation-trap’ (Kongsgaard & Rod, 2018) since limited flexibility during the conversation can hinder a continuous responsiveness and adjustment to the mentee’s wishes and opinions regarding both the conversation process and the content.