Data obtained from dietary intake surveys are often used to estimate the proportion of the population with insufficient (or excessive) intake of certain dietary components. It is generally agreed that the usual or long-run average intake of a nutrient is the appropriate measure of an individual's intake. In this light, assessments of the dietary status of the population should be based on the distribution of usual intakes for each dietary component. Several methods have been proposed for estimating the distributions of usual intakes from dietary data. The methodology presented by Nusser et al. (1995) is very appealing, since it accounts for all of the attributes of dietary intake data, and explicitly recognizes that when data are collected on consecutive days, it is necessary to incorporate the correlations among intake days into the estimation procedure. In this paper we present a method for obtaining smooth estimates of day-to-day correlations for nutrient intake, and their standard errors, when dietary data are collected on consecutive days. The method consists of two steps. In the first step, direct correlation estimates are obtained by fitting a measurement error model with a second-order autoregressive variance structure. Smooth correlation estimates are then obtained from the direct estimates by means of a two-factor model. Data from the combined 1989-CSFII are used to obtain correlation estimates (and their standard errors) for 26 nutrients, water, and six ratios of dietary components. Estimates were obtained for each of five age-sex groups.