The protection of children in cars is improving with the increasing use of better designed restraint systems. Indeed, when children are correctly restrained in appropriate child restraint systems (CRS) they are sufficiently well protected in moderate frontal impacts. However, the levels of protection afforded in severe frontal impacts and lateral crashes has needed further attention. The CREST project, funded by the European Commission, was initiated to develop the knowledge on the kinematics behaviour and tolerances of children involved in car crashes. The final aim of the project is to propose enhanced test procedures for evaluating the effectiveness of child restraint systems (CRS). The method used in this project was to collect data from accident investigations and from reconstructed crashes in order to determine the physical parameters (measured on dummies) which correspond to various injury mechanisms, and is described in ESV 2001 – paper n°294. This paper presents the activity developed within Workpackage I of CREST, responsible for accident investigations. A presentation of the common methodology used is made. A general description of the data base of 405 documented accident cases in which 628 restrained children are involved is given as well as specific features of the reality of the restraint of children in cars, for example:​ age and size, CRS type, place, and a discussion on misuse and inappropriate use. Injuries in frontal and side impacts are also discussed.