This paper describes a method for predicting the real-world safety performance of a new or modified car design over the whole range of crash severities where injuries occur. With a knowledge of the crash severity distribution, the overall effectiveness of alternative design approaches can be estimated by correlating lab test dummy responses with real-world injury risks at different crash severities. The procedure can be used for selecting the most effective among several design options or, inversely, for establishing a dummy response limit curve corresponding to a desired occupant protection goal. Tentatively applied to car-to-car side impacts, the method turned out to be a practicable and promising evaluation instrument.