This paper analyzes tobacco demand within a discrete choice framework. Using binomial and multinomial logit models with random effects, and an unbalanced panel data set of Norwegian households over a twenty year period, we first consider the decisions a) whether to smoke or not, and b) given the choice is to smoke, whether to smoke hand rolled or manufactured cigarettes. Next, we consider a multinomial logit framework, in which the households choose between no tobacco, only manufactured cigarettes, only hand rolled cigarettes, and a combination of manufactured and hand rolled cigarettes. In this process, we utilize the potential offered by panel data to investigate unobserved heterogeneity, which is crucial for commodities where consumers have different tastes and where users tend to become addicted. Using Maximum Likelihood in combination with bootstrap estimation of standard errors, we find that income and prices influence the 'type of tobacco choice probabilities' at least as strongly as the 'smoking/non-smoking probabilities'. Cet.par., an increase in the price of manufactured cigarettes could lead consumers to switch to hand rolled cigarettes, rather than quit smoking. Socio-demographic variables seem to be at least as important in explaining the discrete aspects of tobacco consumption as income and prices. Finally, we find significant unobserved household specific effects in the smoking pattern.