A Bayesian approach was used to analyse household demand for staples in Nigeria within the framework of a multivariate double-hurdle model to account for censoring emanating from non-participation decisions. We demonstrate how to impose identification constraints in the probit equations and introduce a straightforward way of mapping observed and latent shares in the demand (share) equations to satisfy adding up restrictions. Demands for cereals, beans and tubers are own-price inelastic with values close to unity in the lowest income quintile. Cross-price elasticities indicate demand patterns characterised by a mix of gross substitutability and complementarity relationships among staple food subgroups. Presence of children and adolescents in the household as well as rural-urban and zonal (regional) differences exact significant influence on demand for staple foods. Total expenditure elasticities on staple food subgroups decline with higher income levels in line with Engel's law. Cereals and tubers are all necessary goods in the middle and highest income quintiles. However, they are luxury goods in the lowest income quintile. Our findings suggest that economic growth coupled with targeted interventions such as cash or food stamp transfer programmes are crucial for improved consumption of major staples and nutrition among households.