The roadside crossing judgments of children aged 7, 9, and 11 years were assessed relative to controls before and after training with a computer-simulated traffic environment. Trained children crossed more quickly, and their estimated crossing times became better aligned with actual crossing times. They crossed more promptly, missed fewer safe opportunities to cross, accepted smaller traffic gaps without increasing the number of risky crossings, and showed better conceptual understanding of the factors to be considered when making crossing judgments. All age groups improved to the same extent, and there was no deterioration when children were retested 8 months later. The results are discussed in relation to theoretical arguments concerning the extent to which children's pedestrian judgments are amenable to training.