We derive a joint continuous/censored commodity demand system for panel data applications. Unobserved heterogeneity is controlled for using a correlated random effects specification and a generalized method of moments framework used to estimate the model. While relatively small differences in elasticity estimates are found between a flexible random effects specification and one that restricts the random effect coefficient to be time invariant, larger differences are observed when comparing the flexible model to a pooled cross-sectional estimator. The results suggest the limited ability of such estimators to control for preference heterogeneity and unit-value endogeneity leads to parameter bias.