The paper discusses automotive crash simulation in a stochastic context, whereby the uncertainties in vehicle properties, boundary and initial conditions are taken in to account by means of Monte Carlo simulation techniques. It is argued that, since crash is a non-repeatable phenomenon, qualification for crashw orthinessbased on a single test is not meaningful neither from a physical nor from a conceptual standpoint, and should be replaced by stochastic simulation in which the mentioned uncertainties are taken into account. In addition, a broad spectrum of impact angles and velocities should be considered in each scenario (frontal, rear, side, etc.). The advocated approach, due to the fact that it addresses a sample of the population of the to-be-manufactured cars, instead of a single idealized nominal car, possesses built-in robustness and therefore enables the entire problem to be viewed with a high level of confidence. Finally, it is shown that based on today’s deterministic CAE techniques, validation of numerical models via a single test is impossible.