Development of crash avoidance systems and active safety systems must not be only based on experimental knowledge. The goal is to provide an efficient answer to still unsolved severe real-world car crashes which occur despite enhanced passive safety devices. This requires to know precisely the pre-crash conditions during about 3 to 10 seconds before impact. The paper describes the multidisciplinary systemic approach leading to the comprehensive methodology used in accident reconstruction in order to determine the best scenario, and to assess initial car speeds, paths and events in the different phases of the accident. This has already been carried out for about 400 car crashes with car occupant injuries (including 6% fatal and 10% severely injured). The necessity of collecting data on the spot of the crash scene is highlighted. Three well-trained investigators are involved. The first technician collects vehicle characteristics which could change later like tire pressures or gear position. The second technician focuses on measurements of road-marks and skid-marks, accurate impact point location, final car positions an weather conditions as well. The third technician is a psychologist who makes driver interviews in order to record immediately the story of the pre-crash (perceptions, interpretations and actions). After complementary data collection later on, the mathematical computer reconstruction can start taking into account behavioral driver activities. The basic mechanical equations are used to obtain primarily reliable scenarios for post-crash, crash and pre-crash phases, fitting together geometrical data and testimonies. Some kinematics parameters are estimated by a range of probable values with the Monte-Carlo method. Upon this first step, a more accurate accident simulation can be calculated with a reconstruction software. Results and examples of applications are given in the most common pre-crash situations (curves, straight roads and intersections).