153 3-year-old and 144 8/9-year-old children consumed a challenge drink containing artificial food colour (sodium benzoate) and one of two artificial food additive mixes (A or B) or a placebo mix, to test for effects on childhood behaviour. Among all 3-year-old children, behaviour assessed by a global hyperactivity aggregate was significantly and adversely affected by mix A compared with placebo (effect size 0·20 [95% CI 0·01–0·39], p=0·044), but not by mix B versus placebo. 8/9-year-old children showed a significantly adverse effect when given mix A (0·12 [0·02–0·23], p=0·023) or mix B (0·17 [0·07–0·28], p=0·001). In both populations, results persisted when analysis was restricted to those children consuming ≥85% of drinks with no missing data.