This paper deals with three issues. First, the question of the boundaries of safety science – what is in and what is out – is a practical question that journal editors and reviewers must respond to. I have suggested that there is no once-and-for-all answer. The boundaries are inherently negotiable, depending on the make-up of the safety science community.

The second issue is the problematic nature of some of the most widely referenced theories or theoretical perspective in our inter-disciplinary field, in particular, normal accident theory, the theory of high reliability organisations, and resilience engineering. Normal accident theory turns out to be a theory that fails to explain any real accident. HRO theory is about why HROs perform as well as they do, and yet it proves to be impossible to identify empirical examples of HROs for the purpose of either testing or refining the theory. Resilience engineering purports to be something new, but on examination it is hard to see where it goes beyond HRO theory.

The third issue concerns the paradox of major accident inquiries. The bodies that carry out these inquiries do so for the purpose of learning lessons and making recommendations about how to avoid such incidents in the future. The paradox is that the logic of accident causal analysis does not lead directly to recommendations for prevention. Strictly speaking recommendations for prevention depend on additional argument or evidence going beyond the confines of the particular accident.