Purpose: Primary-care referral to community-based exercise specialists (exercise referral) is common in the UK despite limited evidence of effectiveness. A recent pragmatic randomised trial of the Welsh National Exercise Referral Scheme (NERS), demonstrated promising impacts upon physical activity and mental health. This paper presents a mixed-method process evaluation exploring how outcomes were achieved. Design/methodology/approach: Structured observation, implementer interviews and routine data assessed the extent to which NERS was implemented as intended. Baseline trial data were combined with routine monitoring data for the purposes of profiling uptake and adherence. Semi-structured patient interviews explored processes of change and the emergence of social patterning in responses to the scheme. Findings: NERS offered patients a programme of supervised, group-based discounted exercise. However, motivational interviewing, goal-setting and patient follow-up protocols were delivered poorly. The high degree of professional support was perceived as helping patients to build confidence and assimilate into exercise environments. Patient-only classes provided social contacts, a supportive context and realistic models. Patterning in uptake emerged from access issues, with uptake lower among non-car owners. Adherence was poorer among mental health patients, younger patients and those who were least active prior to referral to NERS. Originality/value: In practice, although the NERS RCT demonstrated positive impacts on physical activity and mental health, process evaluation data indicate that the intervention was not entirely delivered as intended. Mixed-method process evaluation served crucial functions in understanding implementation and functioning, offering insights into the roles of professional support and exercise classes in promoting activity and mental health, and the emergence of social patterning in responses to an ERS.