The CRASH3 computer program, a well known and useful tool in accident reconstruction, is shown to be innaccurate by comparison with car-to-car crash test data. Claims for accuracy of about 10 percent cannot be validated. Both the impact model and the damage only model yield results which are in error. Cases involving error well in excess of 20 percent are demonstrated. These inaccuracies are due primarily to the omission of terms in the formulation of the energy equation and to the sensitivity of the solution to the input estimate of principle-direction-of-force.