The authors measured intra-individual day-to-day variation and tracking of nutrient intakes among 181 preschool children (53% male, 45-60 months of age at baseline, 93% Hispanic) recruited through a hospital-based pediatrics practice in New York City. From 1986 to 1989, 24-hour dietary recalls were administered on seven occasions (four times in year 1 and three times in year 3) to the children's mothers. Median follow-up (midpoint of year 1 to midpoint of year 3) was 19.3 months. The reliability of estimates (intraclass correlation coefficients) of energy and nine nutrients obtained from a single administration of the dietary recall ranged from 0.15 to 0.38. Based on unadjusted nutrient intakes, 33.1-55.6% of children in the top quintile of intake at year 1 remained in the top quintile, and 58.3-83.3% in the top two quintiles, at year 3. Of the children in the lowest quintile at year 1, 27.8-50.0% were in the lowest quintile and 55.6-80.6% in the lowest two quintiles at year 3. Consistency of classification decreased when intakes were adjusted for energy intake. Correlations between mean energy and unadjusted nutrient intakes at year 1 and year 3 ranged from 0.27 to 0.45. When energy intake was controlled, correlations decreased for most but not all nutrients (range, 0.09-0.59). Correction of correlations of energy and energy-adjusted nutrients for residual intra-individual variation yielded correlations between mean intakes at year 1 and year 3 in the range 0.15 to 0.71. These data indicate that despite considerable residual intra-individual day-to-day variability of dietary intakes there is substantial tracking of underlying diets among preschool children over a 19-month period.