Abstract
Hazard identification is a major scientific challenge, notably for environmental epidemiology, and is often surrounded, as the recent case of glyphosate shows, by debate arising in the first place by the inherently problematic nature of many components of the identification process. Particularly relevant in this respect are components less amenable to logical or mathematical formalization and essentially dependent on scientists' judgment. Four such potentially hazardous components that are capable of distorting the correct process of hazard identification are reviewed and discussed from an epidemiologist perspective: (1) lexical mix-up of hazard and risk (2) scientific questions as distinct from testable hypotheses, and implications for the hierarchy of strength of evidence obtainable from different types of study designs (3) assumptions in prior beliefs and model choices and (4) conflicts of interest. Four suggestions are put forward to strengthen a process that remains in several aspects judgmental, but not arbitrary, in nature.