This paper proposes several methods for computing precise confidence intervals or evaluating the precision of some statistics related to individual food consumption, based on complex household food survey datas. We show how it is possible to obtain asymptotic confidence intervals for non-linear functionals thanks to the delta method and the notion of Hadamard differentiability. However asymptotic confidence intervals may not be very precise and to not take into account the dissymetries of the statistics or the underlying distributions. We develop two different methods based on resampling ideas to obtain precise confidence intervals. The first one is a transposition of the weighted bootstrap to survey sampling. The second uses the universal properties of subsampling and extrapolation methods to obtain rapidly accurate results. We compare and apply these methods to the construction of confidence intervals for means, fractiles, dispersion indexes of individual food consumptions (with or without null consumptions). We apply these methods to several products from the 1994 Secodip french panel.