Objective In order to fully reconstruct the accident by utilizing pedestrian injuries information gained from the car-pedestrian collision, a new method based on finite element simulation and genetic neural network to deduce the car-pedestrian collision parameters in reverse is proposed. Methods Crash simulations from different contact angles (back, left, front, right) at different impact speeds (25, 40, 55 km/h) were conducted by using Hyperworks and LS-DYNA, so as to obtain the head injury criterion (HIC) value and the maximum velocity of the thoracic wall. According to the criteria of injury biomechanics, the severities of the pedestrian head and thorax and corresponding injury locations were analyzed and set as predictors, and the predictive values of collision parameters were then acquired by using genetic neural network. Finally, this method was verified by two car-pedestrian accidents with the video and exact collision parameters. Results For both cases of the car-pedestrian accidents, the car speeds at the collision of pedestrian were 54 and 49 km/h, respectively, and the car-pedestrian contact angles were both 180°. While according to the pedestrian injuries information, the predictive values of the car speeds at the collision of pedestrian were 51 and 43 km/h, and the predictive values of the car-pedestrian contact angles were 184° and 169°, respectively. The reconstruction accuracies of two cases were 0.94 and 0.88. Conclusions The proposed method in the study can be used to predict car-pedestrian collision parameters efficiently and accurately by utilizing the pedestrian injuries information, which provides a new method for cause analysis and responsibility recognition, as well as theoretical references for the treatment and protection of head and thoracic injuries occurred in the car-pedestrian collision.