Background: Social media use is ubiquitous among adolescents, yet its relationship with mental health may depend on why it is used. This study distinguishes between avoidant-escapist and social-recreational motives for social media use and examines their associations with mental health, wellbeing, and resilience in a large cohort of UK schoolchildren.Methods: Cross-sectional data were drawn from the BrainWaves study (N = 6820, Mage = 16.72). Social media motives were assessed using 10 items and derived using exploratory factor analysis (EFA) into two latent factors: avoidant-escapist (boredom, addiction, distraction, procrastination) and social-recreational (socialising, communication, entertainment, observation) use. Confirmatory factor analysis (CFA) was used to estimate factor scores, which were examined in linear regression models. Structural equation modelling (SEM) was then applied to estimate associations with anxiety, depression, wellbeing, and resilience, adjusting for gender and deprivation.Results: In linear regression analyses, avoidant-escapist social media use was associated with higher anxiety and depressive symptoms, whereas social-recreational use was associated with lower levels of both outcomes. No interaction between gender and social media motives was observed. The SEM was a good fit to the data and showed that avoidant-escapist use was associated with higher anxiety and depressive symptoms, and lower wellbeing and resilience. In contrast, social-recreational use was associated with lower anxiety and depressive symptoms, and higher wellbeing and resilience. Conclusions: The associations between social media use and adolescent mental health differ by underlying motives. Avoidant-escapist use is linked to poorer outcomes, whereas social-recreational use is associated with more favourable outcomes. Focusing on why adolescents engage with social media, rather than overall use alone, is necessary for understanding risk and informing interventions. However, given the cross-sectional design, causality cannot be inferred, and associations may be bidirectional.