The sites and routes that children of different ages considered to be safe to cross the road were investigated. In Expt 1, children aged 5, 7, 9 and 11 years were instructed to choose ‘the safest’ crossing sites and routes to specified destinations. The results showed a gradual developmental shift with safer, more adult‐like choices appearing with increasing age. Five‐ and 7‐year‐olds exhibited only a rudimentary selection procedure, choosing the most direct route as safest and showing a marked lack of awareness of the dangers posed by nearby roadside obstacles or other visual restrictions. In a further experiment, 5‐year‐olds were individually trained in finding safe places to cross. Training took place either in the real road environment or using a tabletop model of a traffic environment. A series of pre‐ and post‐tests enabled the effectiveness of the training to be assessed. Substantial improvements following training were obtained in both groups. No differences were found between the two training methods. Though performance fell somewhat over the two months following training, trained children still outperformed their untrained peers eight months after the programme ended. The implications for road safety educ