We investigated marijuana, alcohol and tobacco consumption using micro-unit data from the Australian National Drug Strategy Household Surveys. We estimated a multivariate probit (MVP) model to allow for correlations across participations of different drugs and a sequential model to study separately the determinants of participation and the levels of consumption. The MVP results indicate significant and positive correlations across all three drugs through unobservable characteristics, with the correlation coefficient between marijuana and tobacco being the highest. The MVP approach allows for better prediction of conditional and joint probabilities, providing valuable information for policy makers in a multidrug framework.