The demand for food and beverages is estimated within a three-stage demand model. The separability structure of the model is checked by nonparametric tests. Some generalized axiom of revealed preference (GARP) violations are detected in one of the subsystems. However, they are removed by small adjustments in the quantities of fish, and the violations are interpreted as results of measurement errors. The almost ideal demand system is used in the static and a dynamic version. The results of various specification and misspecification tests suggest that the static version performs poorly as compared with the dynamic version. Norwegian demand elasticities for disaggregate food commodities have rarely been estimated within a system framework, so the results are of intrinsic interest. The elasticities estimated by using the dynamic model are of the expected signs and reasonable magnitudes. The values are stable over time for most commodities. Elasticities estimated within a subsystem are conditional on the goods included in that system, and they may differ from the more policy relevant unconditional elasticities estimated within a system including all goods. Adjustment formulas are used to approximate the unconditional elasticities from the estimated conditional elasticities. There are considerable differences between the numerical values of the conditional and unconditional elasticities for several of the foods. The unconditional own-price elasticities are in the interval -0.20 to - 0.89. The own-price elasticities for hot drinks and for milk are most inelastic. The unconditional expenditure elasticities for food-away-from-home, fish, soft drinks, and alcoholic beverages are above one, while the expenditure elasticity for hot drinks is about zero. Published by Elsevier Science B.V.