PLAN-A is a cluster randomised controlled trial of a peer-led physical activity intervention which uses peer supporters to increase the physical activity of 13–14-year-old girls in the UK. This paper uses latent class analysis to identify classes in the whole study population and investigate how those selected as peer supporters in PLAN-A were drawn from different social groups. We identified five classes of girls, based on psychosocial variables (self-esteem, physical activity self-efficacy, motivation, physical activity values among friends and peer support for physical activity (PA) and physical activity behaviour variables (average minutes of weekday MVPA, sedentary time and screen viewing). Peer supporters were similar to the whole study population in terms of overall demographics, but were drawn unequally from the five classes. In addition, there was considerable variation in the distribution of peer supporters between schools. The selection of peer supporters is an integral component of peer-led interventions and should be explored and linked to underlying theory to understand the characteristics of those recruited. However, demographic representativeness is not necessarily the aim, and simple reporting of overall demographic comparisons may mask important differences within subgroups.