This paper has four objectives. Firstly, modifications are recommended and presented to a well known set of US Army test data representing eleven tests that involve destruction of buried concrete box structures by ground transmitted shock waves. The resultant database is available for failure mode analysis. Secondly, a new approach forfailure mode analysis based on pattern recognition techniques to the structural engineering community is introduced. Thirdly, a comparison is presented of the results of preprocessing, feature extraction, and cluster analysis for failure modes to several previous studies. The objective here is to inform the structural engineering community about data interpretations that can be inferred using this approach; the authors do not claim to have a conclusive or definitive classification of failure modes in these data. Finally, the importance of a priori decisions about data analysis during the planning stages of physical experiments is emphasized by pointing out the many statistical and other uncertainities that are associated with the data being discussed.