A study was performed to determine the utility of the Willett semiquantitative food frequency questionnaire for assessing the habitual diets of preschool children. Children (n = 224) were recruited mainly through a New York City hospital-based pediatric practice during 1986-1987. The children's ages at baseline were 44-60 months; 50% were male, and 91% were Hispanic. Over a 12-month period, the Willett food frequency questionnaire was administered twice to each child's parent, and a 24-hour dietary recall was conducted four times. For energy and eight nutrients, group mean intakes derived from food frequency questionnaires were 1.4-1.9 times higher than those from 24-hour recalls. Group mean estimates of nutrient density of total and saturated fat, potassium, and calcium did not differ between the two methods. Correlations between methods for crude nutrient intakes (unadjusted for energy consumption) ranged from 0.16 (polyunsaturated fat in boys) to 0.60 (potassium in girls). Correlations generally decreased when intakes were adjusted for energy consumption. Adjustment for energy intake and residual intraindividual variability yielded correlations of 0.48 for total calories, 0.35 for total fat, and 0.37 for saturated fat. For intake of energy and nine nutrients, of those children classified into the highest quintile by dietary recall, 28.9-40.9% were so classified by the Willett questionnaire, and 48.9-68.9% were classified into the highest two quintiles. When data were expressed as nutrient densities, agreement was high for potassium and calcium and fair for saturated fat, cholesterol, and protein. The moderately low consistency of nutrient intake estimates across dietary assessment methods in this study may be due, in large part, to residual intraindividual variability in both the recall data and the food frequency data.