This paper explores gender peer effects heterogeneity in adolescent Body Mass Index (BMI). We propose a utility-based non-cooperative social network model with effort technology. We allow the gender composition to influence peer effects. We analyze the possibility of recovering the fundamentals of our structural model from the best-response functions. We provide identification conditions of these functions generalizing those of the homogeneous version of the model. Extending Liu and Lee [2010], we consider 2SLS and GMM strategies to estimate our model using Add Health data. We provide tests of homophily in the formation of network and reject them after controlling for network (school) fixed effects. The joint (endogenous plus contextual) gender homogeneous model is rejected. However, we do not reject that the endogenous effects are the same. This suggests that the source of gender peer effects heterogeneity is the contextual effects. We find that peers’ age, parents’ education, health status, and race are relevant for the latter effects and are gender-dependent.