The aim of the present study was to determine the errors in the conceptualization of portion size using photographs. Male and female volunteers aged 18-90 years (n 136) from a wide variety of social and occupational backgrounds completed 602 assessments of portion size in relation to food photographs. Subjects served themselves between four and six foods at one meal (breakfast, lunch or dinner). Portion sizes were weighed by the investigators at the time of serving, and any waste was weighed at the end of the meal. Within 5 min of the end of the meal, subjects were shown photographs depicting each of the foods just consumed. For each food there were eight photographs showing portion sizes in equal increments from the 5th to the 95th centile of the distribution of portion weights observed in The Dietary and Nutritional Survey of British Adults (Gregory et al. 1990). Subjects were asked to indicate on a visual analogue scale the size of the portion consumed in relation to the eight photographs. The nutrient contents of meals were estimated from food composition tables. There were large variations in the estimation of portion sizes from photographs. Butter and margarine portion sizes tended to be substantially overestimated. In general, small portion sizes tended to be overestimated, and large portion sizes underestimated. Older subjects overestimated portion size more often than younger subjects. Excluding butter and margarine, the nutrient content of meals based on estimated portion sizes was on average within +/- 7% of the nutrient content based on the amounts consumed, except for vitamin C (21% overestimate), and for subjects over 65 years (15-20% overestimate for energy and fat). In subjects whose BMI was less than 25 kg/m2, the energy and fat contents of meals calculated from food composition tables and based on estimated portion size (excluding butter and margarine) were 5-10% greater than the nutrient content calculated using actual portion size, but for those with BMI 30 kg/m2 or over, the calculated energy and fat contents were underestimated by 2-5%. The correlation of the nutrient content of meals based on actual or estimated portion sizes ranged from 0-84 to 0-96. For energy and eight nutrients, between 69 and 89% subjects were correctly classified into thirds of the distribution of intake using estimated portion size compared with intakes based on actual portion sizes. When 'average' portion sizes (the average weight of each of the foods which the subjects had served themselves) were used in place of the estimates based on photographs, the number of subjects correctly classified fell to between 60 and 79%. We report for the first time the error associated with conceptualization and the nutrient content of meals when using photographs to estimate food portion size. We conclude that photographs depicting a range of portion sizes are a useful aid to the estimation of portion size. Misclassification of subjects according to their nutrient intake from one meal is reduced when photographs are used to estimate portion size, compared with the use of average portions. Age, sex, BMI and portion size are all potentially important confounders when estimating food consumption or nutrient intake using photographs.