We investigate the relationship between social media use and emotional and behavioural outcomes in adolescence using data from a large and detailed longitudinal study of teenagers from the UK. To the best of our knowledge, this is the first study in economics to analyse the effect of social media use on adolescents' mental health. We use individual fixed effects, propensity score matching and treatment effects with Inverse Probability Weighted Regression Adjustment, controlling for a rich set of children's and family's characteristics and using comprehensive sensitivity analyses and tests to assess the potential role of unobserved variables. Our results show that prolonged use of social media (more than 4 hours per day) is significantly associated with poorer emotional health and more behavioural difficulties, and in particular decreased perception of self-value and increased incidence of hyperactivity, inattention and conduct problems. However, limited use of social media (less than 3 hours per day) has some positive effect on peer relationships.