A panel of households is used to test the rational addiction model of Becker et al. (1994). These data raise problems of measurement errors, censoring, and unobservable heterogeneity. We use sample separation information to exclude those households who never purchase tobacco. To deal with the remaining zeros we compare specifications based on infrequency of purchase and on censoring. GMM and system-GMM are used to deal with errors-in-variables and unobservable heterogeneity. Within-groups two-step, within-groups three-step GMM and Minimum Distance methods are used to allow for censoring. There is evidence that the rational addiction specification is sensitive to unobservable heterogeneity and censoring.