Methods for assessing change in the habitual diet of children are essential for diet intervention studies as well as clinical management. Food frequency questionnaires are a potential alternative to recall and record methods, which require multiple days of data collection for stable individual estimates of habitual intake. Over 3 years, we studied 173 children (93% Hispanic; baseline age 44-60 months) in New York City. We obtained dietary data by interviewing the child's mother. We calculated intakes of nine nutrients, expressed as nutrient densities, as the mean of two administrations of the Willett food frequency questionnaire and the mean of three administrations of the 24-hour dietary recall in years 1 and 3. The two methods consistently estimated the direction of change in mean nutrient density of total and polyunsaturated fat, cholesterol, carbohydrate, and potassium. Changes in nutrient density assessed by the two methods correlated poorly (r < or = 0.15) for all nine nutrients. Cross-classification analysis also showed no relation between change assessed by recall and food frequency methods. Lack of between-person variability could not explain the low correlations, as individual changes in nutrient density were large.