We derive necessary and sufficient conditions for a finite data set of price and demand observations to be consistent with an additively separable preference. We do so without imposing concavity on any of the subutility functions or convexity of the budget set a priori, thereby generalizing earlier results. Our simple and intuitive lattice test easily accommodates departures from rationality, or errors, which subsequently facilitates a rich empirical analysis. We apply our econometric techniques to the food consumption of a panel of British households. The primary empirical finding is that additive separability has considerable success in explaining the data.