Real world crash investigation constitutes the source of all safety research. It gives information on priorities in terms of body segments to protect, crash configurations and impact severities. It can also provide injury tolerances (for instance thoracic injury risk curves) and allows the verification of the efficiency of protective devices.

However, some limitations exist, in particular the precision of information about crash conditions. Experts do good work on defining crash characteristics, using databases, reconstruction software and their own experience. However, even with the best diagnosis, occupant position before the crash is generally still uncertain. This lack of precision can be somewhat compensated by the number of cases which far exceeds the number of biomechanical cases. Furthermore, the data comes from living people, who are after all our subject of interest.

Nevertheless, biomechanical investigations allow compensation of the shortcomings of crash investigations. Probably the most important contribution from biomechanics is the definition of human behavior, without which crash investigations remain only observation. As a matter of fact, human substitutes are needed to provide engineers with tools for improving safety. Only biomechanical research is able to give specifications for the development of such substitutes. Both behavior and injury criteria are important to define. Information on Human behavior is needed in order to mimic injury mechanisms, criteria are needed to measure them. If one or the other is missing, injury protection cannot be evaluated. The next step is the definition of injury risk curves. In this case, either crash investigation or biomechanics can contribute to their definition, the precision and the control of boundary conditions being the strength of biomechanics, the number of cases and the type of subject (living people) being the strength of crash investigation.

So, the use of the two disciplines together provides tools for developing safety systems and allows the evaiuation of their efficiency. They allow also highlighting mechanisms which are different from those experienced in standard procedures with current dummies and the improvement of the behavior of these dummies in order for them to be more humanlike. It also shows that dependence on rating systems is not enough and that in some cases looking at real life is necessary to improve protection.