To evaluate factors that affect the reproducibility of a semi-quantitative food frequency questionnaire used in a large prospective study we compared an extended 99-item questionnaire with a shorter, 61-item form completed by 1497 women with an interval of nine months between. Correlation coefficients for individual items assessed by the two questionnaires were highest for beverages (Spearman r = 0.70). For other foods, coefficients ranged from 0.60 to 0.70 for items eaten frequently (or habitually), to values between 0.34 and 0.45 for foods, such as sweet potatoes and ready made pie, that were eaten less frequently. For food items, the correlation between mean frequency of consumption and the reproducibility coefficient (Spearman r) was 0.51 (p less than 0.01), formally confirming that the reproducibility of measurements was positively associated with frequency of use. Pearson correlation coefficients for calorie-adjusted intakes of nutrients between the two questionnaires ranged from 0.40 for trans-fatty acids to 0.71 for vitamin E (including supplements). These correlation coefficients did not vary materially between subjects in different categories of smoking status or tertiles of age or relative weight. Moderate alcohol use had minimal effect on correlation coefficients, but reproducibility was slightly reduced among heavier drinkers. These data indicate that this self-administered dietary questionnaire can provide reproducible information about individual food and nutrient intakes which is not altered materially by age and a number of important health habits.