Observers seated in a car were required to predict where they would meet an oncoming car (Exp. I). Meeting point predictions were found to be systematically biased in the direction of the midpoint between cars. The error was proportional to the distance between midpoint and meeting point. A reduction in the errors occurred when the observers were allowed to check their predictions (Exp. II). Perfect agreement was found to hold between subjective midpoint judgments (Exp. III) and meeting point predictions, the former showing a small but obvious regression toward the meeting point.