This paper uses non-parametric regression with smoothness induced by imposition of a roughness penalty to estimate time-dependent relationships between age and intakes of energy, total fat and saturated fatty acids employing data on household food acquisitions from a continuous survey covering the period 1975-94. A discrete analogue of the bivariate thin plate spline smoother is used to smooth the estimates across age and time. With age and calendar time-specific estimated intakes in hand it is possible to track changes with age within cohorts. Of particular interest in this application are estimates of the change through time and across age in measures of dietary composition which figure in many governments' dietary targets such as the proportion of energy derived from fat. The method used here can be applied in any situation in which household aggregate data are available on consumption of private goods and there is a need to model the demands of household members or to control for household composition, opening the way to the construction of individual pseudo-panels from household aggregates.