A growing trend in demand analysis during the last two decades is the use of household survey data. Detailed demographic information collected in these surveys allows treatment of heterogeneous preference and the typically large sample also allows estimation of large demand system that are otherwise not possible with aggregate time series. However, the use of household-level data is complicated by the censoring of the dependent variable especially for systems with disaggregated commodity definitions. To overcome the numerical problem of evaluating truncated multi-dimensional error term distributions, a Quasi-maximum likelihood method is used to estimate a censored 9-commodity demand system for a sample of urban Mexican households. The impacts of changes in price and expenditures are quantified as are the impacts of alternative household compositions evaluated via the use of an endogenously determined equivalence scale.