The reconstruction approach is typically a process in which the effects (positions of vehicles at rest, various traces, permanent deformation, etc.) are examined to determine the causes (positions, velocities, acceleration of vehicles and occupants, etc.), and is in general numerically ill-conditioned, with the consequence that minor disturbance in the input data can produce broad variations in the final results. Since the input data and the parameters necessary for utilizing the physical models are known or estimated only with a certain degree of uncertainty, it follows that, for a given accident, substantially different scenarios may be envisaged. In this article, sensitivity to uncertainty in the input data for reconstructing traffic accidents is analysed. The effect of redundant data on calculation is analysed and parameters useful for identifying the data having the greatest effect on error propagation are indicated, for the purpose of reducing dispersion in the results.