Recent findings from real-world accident data have shown that fatality risks for pedestrians are substantially lower than generally reported in the traffic safety literature. One of the keys to this insight has been the large and random sample of car-to-pedestrian crashes available in the German In-Depth Accident Study (GIDAS). Another key factor has been the proper use of weight factors in order to adjust for outcome-based sampling bias in the accident data. However, a third factor, a priori of unknown importance, has not yet been properly analysed. This is the influence of errors in impact speed estimation. In this study, we derived a statistical model of the impact speed errors for pedestrian accidents present in the GIDAS database. The error model was then applied to investigate the effect of the estimation error on the pedestrian fatality risk as a function of car impact speed. To this end, we applied a method known as the SIMulation-EXtrapolation (SIMEX) method. It was found that the risk curve is fairly tolerant to some amount of random measurement error, but that it does become flattened. It is therefore important that the accident investigations and reconstructions are of high quality to assure that systematic errors are minimised and that the random errors are under control.