In this paper we explore the statistical properties of the distributions of consumption expenditures for a large sample of italian households in the period 1989–2004. Goodness-of-fit tests show that household aggregate (and age-conditioned) consumption distributions are not log-normal. Rather, their logs can be invariably characterized by asymmetric exponential-power densities. Departures from log-normality are mainly due to the presence of thick lower tails coexisting with upper tails thinner than gaussian ones. The emergence of this irreducible heterogeneity in statistical patterns casts some doubts on the attempts to explain log-normality of household consumption patterns by means of simple models based on gibrat’s law applied to permanent income and marginal utility.