The doubly-labelled water (2H218O) method was used to measure total energy expenditure (TEE) in ten non-pregnant, non-lactating (NPNL), six pregnant (P) and fourteen lactating (L) women in a rural Gambian community. Measurements were made on free-living subjects at a period of peak energetic stress when high agricultural work loads coincided with a hungry season to induce moderately severe negative energy balance. TEE averaged 10.42 (SD 2.08) MJ/d, equivalent to 1.95 (SD 0.38) times resting metabolic rate (RMR). The energy cost of physical activity plus thermogenesis, derived as TEE – RMR, averaged 4.94 (SD 1.96) MJ/d. Expressed per kg body-weight (103 kJ/kg per d) this component of expenditure was 2.5 times greater than comparative values from inactive, affluent women studied previously (39 kJ/kg per d). Estimated energy intake (EI) in a subset of the women (n 13) was only 4.80 (SD 1.58) MJ/d, yielding an apparent deficit of 6.08 MJ/d between EI and TEE. Weight changes suggested that endogenous fat oxidation accounted for only about 0.85 MJ/d, leaving an unexplained difference of over 5 MJ/d. Critical analysis of possible errors suggests that the new doubly-labelled water method has provided the most reliable estimates and that the estimates of EI were substantially in error. This finding has important consequences for other food intake studies.