There is a debate in scientific literature about the effectiveness of a peer-led approach to anti-bullying interventions. In order to understand which circumstances and for whom these approaches work best, the present study was carried out within the NoTrap! anti-bullying program. Using a cluster design, classes were randomly assigned into two different peer educator recruitment strategies: volunteering (N = 500; 48% females; mean age = 13.5 years, ds = 1.3) vs peer nominated (N = 466; 38% females; mean age = 13.9 years, ds = 1.3). Results showed that voluntary peer educators suffered a higher level of victimization, while the nominated ones tended to be more popular and likable. Furthermore, a set of linear mixed-effect models showed that the program was effective in reducing bullying and victimization, and in increasing defending behaviour only in the voluntary recruitment condition. On the contrary, in classrooms under the peer nominated recruitment condition, bullying and victimization remained stable, and defending behaviour increased only for peer educators, but not for their classmates. This implies that the step of peer selection and recruitment must be kept into consideration in developing and validating an intervention, because of its possible impact on the effectiveness of the whole intervention.