A theory is presented of how a driver might visually control his braking. A mathematical analysis of the changing optic array at the driver's eye indicates that the simplest type of visual information, which would be sufficient for controlling braking and would also be likely to be easily picked up by the driver, is information about time-to-collision, rather than information about distance, speed, or acceleration/deceleration. It is shown how the driver could, in principle, use visual information about time-to-collision in registering when he is on a collision course, in judging when to start braking, and in controlling his ongoing braking. Implications of the theory for safe speeds and safe following distances are discussed, taking into account visual angular velocity detection thresholds, and some suggestions are made as to how safety on the roads might be improved.