OBJECTIVES: Whilst cognitive analytic therapy (CAT) is increasingly used as an indirect intervention, there is little evidence of how the approach can be applied to reflective practice. This study sought to develop a process model of cognitive analytic reflective practice (CARP) groups. DESIGN: Constructivist grounded theory approach supplemented by quantitative measures of helpfulness and group cohesion. METHODS: Twenty-four participants, working within four staff teams in a secure children's home, attended four CARP groups over a 1-year period. Sessions were audio recorded, transcribed and checked for model fidelity and then analysed using grounded theory. Theoretical sampling was achieved via conducting focus groups with teams informed by the emergent categories, and using sessional measures of group cohesion and helpfulness to confirm theoretical saturation. RESULTS: The model constructed had three interrelated main categories: 'facilitator processes: establishing a reflective space', 'group processes: widening awareness of the self, others, and system' and 'outcomes: changing relational dynamics and finding exits.' The model was validated by evidence from the measures. CONCLUSIONS: The model offers an empirical understanding of how CAT informs reflective practice through a better understanding of reciprocity in the dynamics of care systems.