This paper discusses some of the very serious problems associated with the use of anthropomorphic dummies in crash testing of automobiles. These problems arise primarily from the non-repeatable behavior of the dummy and from the unknown correlation between it's behavior and that of the human. The combination of these factors leads to scatter in test results and in their interpretation which are well beyond the usual range of engineering practice.The systems approach to crash testing can only be placed on a true scientific basis by increasing our understanding of the dummy/human relationship and by decreasing test scatter. Test programs relating to these problem areas are described. These tests demonstrate that the results obtained are heavily dependent on the data analysis scheme employed.Results obtained by use of the mathematical analog technique are compared to other methods. The analog approach is found to minimize scatter in test results and yield values which compare favorably with our limited human tolerance knowledge. Results of dummy head impacts against windshields are shown to approximate theoretical considerations when the brain analog technique is employed.The study concludes that the organ analog approach appears to analyze dummy test results in a more meaningful manner than other current techniques; however our knowledge of human tolerance as well as of dummy construction will require considerable advancement before safety systems testing is upgraded from an art to a science.