The accuracy of the CRASH computer program was evaluated in terms of its ability to estimate impact speed. A comparison of the results from CRASH2, CRASH3 and EDCRASH were presented along with measured results from twelve staged collisions. Statistical analysis of these results revealed the impact speeds estimated by these CRASH programs were within −6 to +7 percent of the combined impact speeds at a 95 percent level of confidence. Using EDCRASH's extended features to optimize the input data improved the range to within −3 to +3 percent of combined impact speeds. An example was used to illustrate the use of the confidence intervals to estimate the expected range of impact speed for a given reconstruction. The results for oblique collisions were found to be significantly more accurate than the results for collinear collisions. Runs without the trajectory simulation were found to be significantly more accurate than runs with the trajectory simulation, although good steer angle data greatly improved the results. Accuracy for delta-V and separation velocity were not assessed because of errors in the measured (test) values.