Abstract Aims To quantify the relationship between alcohol and violence with increasing age. Methods Data were from The National Longitudinal Study of Adolescent to Adult Health (ADD Health) of 20,386 people representative of the US population. Mean age at the first wave of interviews was 16.2 years, with subsequent interviews mean of 1, 6.3 and 12.9 years later. We used random-effects models and predictive marginal effects of the association between varying quantities of alcohol consumption and violence while controlling for possible confounders. Results Violence was reported by 19.1% of participants at wave I but just 2.1% at wave IV. The random-effects model showed that consuming 1–4 drinks on each occasion was associated with a modest increase in risk of violence in both males (odds ratio (OR) 1.36, 95% CI 1.13–1.63) and females (OR 1.33, 95% CI 1.03–1.72). For consumption of five or more drinks on each occasion, the risk remained similar for females (OR 1.40 (0.99–1.97)) but increased considerably for males (OR 2.41 (1.96–2.95)). Predictive marginal effects models confirmed that violence rates decreased with age. Conclusions Alcohol is most strongly linked to violence among adolescents, so programmes for primary prevention of alcohol-related violence are best targeted towards this age group, particularly males who engage in heavy episodic drinking. We have analysed data from a large longitudinal, nationally representative sample to investigate the longitudinal relationship between alcohol and violence. We found that alcohol is more strongly linked to violence among adolescents than in adults, particularly males rather than females who engage in heavy episodic drinking.