BACKGROUND: Increasing pressure on emergency services has led to the development of different models of care delivery including GPs working in or alongside emergency departments (EDs), but with a lack of evidence for patient safety outcomes. AIM: This study aimed to explore how care processes work and how patient safety incidents associated with GPs working in ED settings may be mitigated. DESIGN AND SETTING: Realist methodology with a purposive sample of 13 EDs in England and Wales with different GP service models. The study sought to understand the relationship between contexts, mechanisms, and outcomes to develop theories about how and why patient safety incidents may occur, and how safe care was perceived to be delivered. METHOD: Qualitative data were collected (observations, semi-structured audio-recorded staff interviews, and local patient safety incident reports). Data were coded using 'if, then, because' statements to refine initial theories developed from an earlier rapid realist literature review and analysis of a sample of national patient safety incident reports. RESULTS: The authors developed a programme theory to describe how safe patient care was perceived to be delivered in these service models, including: an experienced streaming nurse using local guidance and early warning scores; support for GPs' clinical decision making, with clear governance processes relevant to the intended role (traditional GP approach or emergency medicine approach); and strong clinical leadership to promote teamwork and improve communication between services. CONCLUSION: The findings of this study can be used as a focus for more in-depth human factors investigations to optimise work conditions in this complex care delivery setting.