Day-to-day variability in dietary intake makes it difficult to measure accurately the "usual" intake of foods and nutrients. The objectives of the present study were to estimate within- and between-subject variability for foods and nutrients by adjusted and unadjusted models and to assess the number of days required to assess nutrient and food group intakes accurately by two different methods. Adult men and women aged 18-65 y (n = 1543) in the Food Habits of Canadians Study provided a 24-h recall. A repeat interview was conducted in a subsample to estimate components of variability. Within- and between-subject variability were determined by mixed model procedure (crude and adjusted for age, gender, education, smoking, family size and season). The number of days required to obtain various degrees of accuracy was ascertained by two methods, one that uses the variance ratio for groups and one that considers within-subject variability alone for individuals. Variance ratios were higher using the adjusted compared with the unadjusted method (e.g., for men, energy 1.07 vs. 0.49). More days were required to reflect usual intake with accuracy using the adjusted model (energy 5 vs. 2 d), indicating the need to control for confounders to obtain reliable estimates of intakes.