In this paper I analyze adolescent peer effects on cigarette consumption while considering the 'popularity' of peers. The analysis is based on AddHealth data, a four wave panel survey representative of American high-school students. The data include the social network of each school, which we use to measure peers' popularity from network centrality measures, in particular weighted-eigenvector centrality. We use lagged peers' behavior at the grade level to alleviate potential homophilic confounds, and we include school fixed effects to control for contextual confounds. We find that most of the aggregate peer effects regarding cigarette smoking come from the smoking propensity of the 20% most popular kids, suggesting a mediation from social status. This effect persists seven and thirteen years later (wave 3 and 4 of the data). Indeed, the smoking propensity of the bottom 80% seems to have a negative influence on the probability of smoking in the long run (wave 3 and 4). These results hint to the importance of knowing not only the smoking propensity within a school but also the place of the smokers within the social hierarchy of the school.