This paper uses latent class analysis (LCA) to estimate the amount of underreporting on the BLS Consumer Expenditure Quarterly Survey (CEQ). Specifically, it models underreporting in a given commodity category by those reporting a purchase of any item within that category. This work builds on work presented at ASA last year and is based on an analysis of micro-level, procedural indicators. Data from the CEQ for the years 1996 to 2003 are used in the analysis. A variety of LC models are used to evaluate observed expenditure reporting patterns in several commodity categories. Model covariates include characteristics of the interview, the respondent, and the household. Best-fitting models are determined from well-known statistical tests and subjective diagnostics developed by the authors. Issues related to correlated classification errors are discussed, and methods for combining and evaluating estimates of underreporting are developed.