A brief account is given of mental retardation and intelligence testing in unirradiated human populations, without which it is not possible to judge the 1991 ICRP recommendations relating to mental impairment. Their basis seems to be an inadequate appreciation of aspects of mental retardation, of intelligence testing and their correlation. The dose-response used by ICRP (1991) for the reduction of IQ by irradiation in utero has no radiobiological basis because IQ values are derived from intelligence test scores by transforming the scale of measurement. It is also defective because it assumed that IQ is distributed normally whereas this is so only in normal school children, not in a population in general including retarded persons. There seems good evidence for a substantial threshold of dose for both reduction in IQ and increase in severe mental retardation (SMR). The four prenatally irradiated bomb survivors with SMR and intrauterine dose in the dose range 1-49 cGy were not intelligence tested, so the relation between SMR and IQ in that practically important dose range cannot be examined directly. SMR is a deterministic phenomenon, so it is not expected to occur unless dose exceeds a threshold. The threshold doses for SMR based on linear dose-responses using ungrouped doses were 46 and 55 cGy DS86 intrauterine dose (Otake et al 1987). The threshold 50 cGy derived here for reduction in IQ is closely similar. Mild mental retardation has not been reported as a diagnosis in bomb survivors exposed in utero.