Five-year-old children were trained in road-crossing skill using a new method which allows them to act safely in relation to vehicles on a normal road. The children learned to time their crossings of a "pretend road" as if the vehicles were on this, rather than on the adjacent road. A previous study, using a single lane of traffic, showed that many children performed well in this simulation with minimal instruction, but that five-year-olds were generally less proficient than older children. In the present study, the method was extended to the more realistic case of two-way traffic, and training programmes for five-year-olds were assessed. (In addition, the performances of adults in the two-way pretend task and in actually crossing the road were compared; the results confirmed the validity of the simulation. After a few sessions of guided practice, the children's efficiency in making use of gaps by setting off promptly after a vehicle had passed improved markedly; in single-lane crossing they reached almost adult standard. In both single-lane and two-way crossing they became as successful as adults in completing their crossings before a second vehicle passed and they took account of the duration of the gap by crossing more quickly when time was short. They remained very cautious, rejecting many adequate gaps. The standard of performance after a few sessions of two-way crossing was maintained over a three week break in training. Practice with one-way traffic did not help them with two-way crossing. In sum, after practice in the simulation, five year olds develop a degree of competence normally shown by older children, whose experience on the roads puts them at lower risk. This suggests that the simulation, which allows children to safely explore and develop their capabilities more fully than when actually crossing the road, would be a valuable addition to road safety programmes.