The addictive nature of tobacco and alcohol suggests that it is more appropriate to model these consumption as a system. We expand Heckman’s sample selection model into a censored system (or multivariate sample selection model) to analyze household tobacco and alcohol consumption. We use a pooled cross sectional data of 77,043 observations from Polish Household Survey data in the period of 2005 to 2008 and apply full information maximum likelihood estimation. Empirical investigation indicates that the decisions to smoke and drink as well as their expenditure levels are indeed, respectively, positively correlated. We examine the effects of demographic, socio-economic factors and outmigration, a special issue in Poland, on the consumption decisions and expenditure on tobacco and alcohol. Findings provide insights for the reduction and prevention of tobacco and alcohol use.