Pedestrian injuries contribute greatly to child morbidity and mortality. Recent evidence suggests that training within virtual pedestrian environments may improve children's street crossing skills, but may not convey knowledge about safety in street environments. We hypothesized that (a) children will gain pedestrian safety knowledge via videos/software/internet websites, but not when trained by virtual pedestrian environment or other strategies; (b) pedestrian safety knowledge will be associated with safe pedestrian behavior both before and after training; and (c) increases in knowledge will be associated with increases in safe behavior among children trained individually at streetside locations, but not those trained by means of other strategies. We analyzed data from a randomized controlled trial evaluating pedestrian safety training. We randomly assigned 240 children ages 7-8 to one of four training conditions: videos/software/internet, virtual reality (VR), individualized streetside instruction, or a no-contact control. Both virtual and field simulations of street crossing at 2-lane bi-directional mid-block locations assessed pedestrian behavior at baseline, post-training, and 6-month follow-up. Pedestrian knowledge was assessed orally on all three occasions. Children trained by videos/software/internet, and those trained individually, showed increased knowledge following training relative to children in the other groups (ps < 0.01). Correlations between pedestrian safety knowledge and pedestrian behavior were mostly non-significant. Correlations between change in knowledge and change in behavior from pre- to post-intervention also were non-significant, both for the full sample and within conditions. Children trained using videos/software/internet gained knowledge but did not change their behavior. Children trained individually gained in both knowledge and safer behavior. Children trained virtually gained in safer behavior but not knowledge. If VR is used for training, tools like videos/internet might effectively supplement training. We discovered few associations between knowledge and behavior, and none between changes in knowledge and behavior. Pedestrian safety knowledge and safe pedestrian behavior may be orthogonal constructs that should be considered independently for research and training purposes.