Context: Direct observation (DO) of residents’ performance, despite the importance that is ascribed to it, does not readily fit in with the practice of postgraduate medical education (PGME); it is infrequent and the quality of observation may be poor in spite of ongoing efforts towards improvement. In recent literature, DO is mostly portrayed as a means to gather information on the performance of residents for purposes of feedback and assessment. The role of DO in PGME is likely to be more complex and poorly understood in the era of outcome-based education. By exploring the possible complexity of DO in workplace learning, our research aims to contribute to a better use of DO in the practice of PGME. Methods: Constructivist grounded theory informed our data collection and analysis. Data collection involved focus group sessions with supervisors in Dutch general practice who were invited to discuss the manifestations, meanings and effects of DO of technical skills. Theoretical sufficiency was achieved after four focus groups, with a total of 28 participants being included. Results: We found four patterns of DO of technical skills: initial planned DO sessions; resident-initiated ad hoc DO; supervisor-initiated ad hoc DO, and continued planned DO sessions. Different patterns of DO related to varying meanings, such as checking or trusting, and effects, such as learning a new skill or experiencing emotional discomfort, all of them concerning the training relationship, patient safety or residents’ learning. Conclusions: Direct observation, to supervisors, means much more than gathering information for purposes of feedback and assessment. Planned DO sessions are an important routine during the initiation phase of a training relationship. Continued planned bidirectional DO sessions, although infrequently practised, potentially combine most benefits with least side-effects of DO. Ad hoc DO, although much relied upon, is often hampered by internal tensions in supervisors, residents or both.