Aims To identify childhood and adolescent predictors of alcohol use and harmful drinking in adolescence and adulthood. Design Longitudinal data from childhood to mid-life from the National Child Development Study (NCDS) were used, including predictors collected at ages 7, 11, 16 years and alcohol outcomes collected at ages 16, 23, 33 and 42 years. Setting The NCDS is an ongoing longitudinal study of a cohort of 1 week's births in Britain in 1958. Participants Childhood and adolescent predictors and alcohol use data from at least one adolescent or adult wave were available from 7883 females and 8126 males. Measurements Social background, family, academic and behavioural predictors measured at ages 7, 11 and 16 years were entered into hierarchical multiple and logistic regressions to predict quantity of alcohol use at ages 16, 23, and 33 years and harmful drinking [i.e. Cut-down, Annoyed, Guilt, Eye-opener (CAGE) questionnaire score] by age 42 years. Findings Previous drinking was controlled in final models to predict change. Drinking was heavier among those with greater childhood and adolescent social advantage (especially females), less harmonious family relationships, more social maladjustment, greater academic performance, less internalizing problems, more truancy and earlier school-leaving plans. Conclusions Alcohol use and problems in adulthood can be predicted by indicators of social background, adjustment and behaviour in childhood and adolescence. Results demonstrate that the early roots of adolescent and adult alcohol use behaviours begin in childhood.