The use of the Part 572 dummy in crash testing is reviewed, noting that the variability of the results obtained from crash tests is unsatisfactorily high. The procedures lack flexibility to check all of the possibly important vehicle impact configurations and we are a long way from establishing reliable biomechanical data to cover the types of injury and population variability that need to be considered. It is concluded that this type of testing is not suitable for legislative purposes.An alternative is to use an integrated regime of simple, controlled subsystem tests calibrated from in-depth accident data. Calibration against accident data takes better account of the variability of the user population than can testing with a fiftieth percentile dummy and bypasses the difficulties inherent in the integrated test. As an example of this method, a previously unpublished study of ECE Regulation Number 12 (FMVSS 203) is used.