Not least due to the new ＇Seveso II Directive＇, Safety Management Systems (SMS) have become a hot topic in the ＇safety business＇. To reveal in a structured way how organisations manage safety at their plants, classification schemes can be applied to identify plant-specific precautions on different hierarchical levels, e.g. from ＇top level＇ equipment reliability to ＇bottom level＇ safety climate. In this paper, such a model is used to classify the accidents reported to the European Commission＇s Major Accident Reporting System (MARS) according to the levels on which failures resulting in actual accidents did occur, and to cross-compare the levels of causation with the overall ＇severity＇ of the accidents. It is analytically shown that similar to 66% of the accidents are caused by latent SMS failures, and that especially those accidents which are caused by failures in the ＇deepest layers of latency＇ have significantly higher ＇severities＇ than accidents caused by more ＇immediate＇ failures. In other words, it is quite likely that the deeper the underlying causes of an accident, the more ＇severe＇ its consequences. Implementing SMS is thus worth the effort.