The aim of the present study was to establish whether the characteristics of members of a large national birth cohort study who submitted diet diaries with implausibly low-energy intake differed from those whose recorded energy intake was more plausible. Survey members (n 1898) recorded their diets in a 7d diary in household measures. Those whose reported energy intake (EI) as 'a fraction of their estimated BMR was less than 1·10, here termed low-energy reporters (LER) but often called under-reporters, constituted 20·6% of the study population. None of the variables describing dietary, smoking or exercise behaviour bore a significant relationship with low EI/BMR (<1·10), neither did those describing region of residence, subjective adequacy of income, current social class, social relations or the social environment of the subjects. Results of logistic regression analysis showed that the only independently significant characteristic for men was higher BMI. In women, in addition to higher BMI, having been overweight or obese as an adult independently, but less significantly, predicted low EI/BMR, while membership as a child of social class III (non-manual), having more children in the household and having a paid job marginally but independently decreased the probability of reporting low EI/BMR. Submission of a diary with EI/BMR < 1·10 7 years earlier in the same survey was an even more powerful predictor of current low EI/BMR than higher BMI in both sexes. The average reported diet-composition of LER was more micronutrient- and protein-rich than that of the others, indicating different dietary, or diet-recording, behaviour in this group of subjects. LER are not a random sample of the survey population, and their characteristics, definable to some extent, put them at risk for lower health status. Although EI/BMR cut-off points can be used to identify LER, the problem of how to use their data is still unresolved.