Crash reconstruction is sometimes used to study injury mechanisms and thresholds, but is often difficult because crash and model parameters are not known precisely. If simulation is used as part of the reconstruction process, then various Design-of- Experiment (DOE) tools may be easily applied to estimate response surfaces of the dependent variable (e.g. head acceleration), to a range of possible crash factors, subject to the validity of the model. This approach relies on the validity of the model’s characteristics over the range of likely crash conditions, meaning that non-linear aspects of the system will often need to be included. The contact between the head of a pedestrian and the hood of a car is an example of a nonlinear contact that is critical to the estimation of the variable of interest: the head impact severity (as measured by linear and angular acceleration or HIC, for example). This paper describes the reconstruction of four pedestrian collisions in which the effects of uncertainties in posture and impact speed on the estimation of a head impact severity were quantified. For each case, physical tests were conducted at lower, middle and upper estimates of head impact speed on a vehicle of the same make and model as the striking vehicle in the collision. The results of these tests were used to define a single non-linear contact characteristic in MADYMO that could reproduce the results of all three impact tests. This contact characteristic was then used in the simulation of the collision to estimate a likely range for the head impact severity.