The subject of this thesis is the development and prototypical realization of a driver assistance system with the purpose of predictive pedestrian protection. The system shall help the driver to avoid collisions with pedestrians by issuing a noticeable warning to the driver prior to a possible collision. If the driver fails to react to the warning and the criticality of the situation increases, the system will initiate an automatic braking intervention in order to prevent or mitigate the collision. One of the main challenges for the system is to correctly estimate the risk of an impending collision. For a successful driver warning it is necessary to issue the warning early enough, enabling the driver to react. Therefore, the time before the collision, at which the decision to warn the driver is made, has to be significantly longer than the reaction time of the driver (up to 2 s). Because of this comparatively long time span it is vital for the system to predict the possible movements of the pedestrian as accurately as possible. This also holds true for the automatic braking intervention. With a high assumption placed upon the movement capabilities of pedestrians, the decision for an automatic braking is not possible up to a few hundred milliseconds before the collision, rendering the benefit of the system comparatively small. Therefore, the first part of the thesis concentrates on the development of a situation analysis approach which considers the movement capabilities of pedestrians. The possible and relevant trajectories of pedestrians in typical accident scenarios are analyzed and contribute to the development of the pedestrian motion model. For this model, a test study is designed and conducted to measure the movement capabilities of different test persons in relevant situations. The results of the study are analyzed and integrated into the model. The prediction of the movement capabilities depends on the current velocity of the pedestrian, as well as the available time and direction of movement, which yields significant improvements of the results in the situation analysis. The results show that a decision for an automatic braking intervention based upon the prediction of an unavoidable collision can be made earlier which leads to a reduction in the collision velocity. The second part of the thesis analyzes the sensor system which is used to recognize pedestrians in front of the vehicle and the impact of this system's errors in the situation analysis. The prototypical realization of the system uses a stereo-vision system in order to detect pedestrians and to measure relevant data, for instance position and velocity of the pedestrian relative to the vehicle. The quality of this data is vital for the system to function, therefore, the implications of erroneous data are analyzed, and the requirements for the relevant input data are derived. For this purpose, a sensitivity analysis with a series of simulations is conducted. The artificial sensor data in the simulations is superimposed by artificial noise in order to determine the acceptable degree of noise for the system. The type of noise depends on the data and is derived from the analysis (theoretical as well as practical) of the stereo-vision system. The thesis is concluded by presenting the test-vehicle and an analysis of the system performance in 50 hours of test driving in urban areas.