Computable General Equilibrium (CGE) models are increasingly being used to project world food markets in order to support forward-looking policy analysis. Such projections hinge critically on the underlying functional form for representing consumer demand. Simple functional forms can lead to unrealistic projections by failing to capture changes in income elasticities of demand. We adopt as our benchmark the recently introduced AIDADS demand system and compare it with several alternative demand systems currently in widespread use in CGE models. This comparison is conducted in the context of projections for disaggregated global food demand using a global CGE model. We find that AIDADS represents a substantial improvement, particularly for the rapidly growing developing countries. For these economies, the most widely used demand systems tend to over-predict future food demands, and hence overestimate future production and import requirements for agricultural products.