The paper approaches the linear no-threshold (LNT) hypothesis, currently used as the basis for recommendations in radiological protection, from the point of view of the radiation mechanism. All considerations of the validity of the LNT hypothesis based on experiment or epidemiology are dismissed because of the impossibility of deriving statistically significant data at very low doses. Instead, the LNT hypothesis is assessed from a consideration of the mechanism of radiation action. The DNA double-strand break is proposed to be the crucial radiation-induced molecular lesion. A trace is made using a series of correlations that link the DNA double-strand break to effects at the cellular level and these cellular effects are linked to the induction of cancer. Multistep modelling of carcinogenesis is used to take the link through to a consideration of radiation risk. It is concluded that, from the point of view of radiation mechanism, at very low doses the LNT hypothesis of radiation action is valid, that is, the risk function has a positive slope from zero dose.