If relative prices of goods within a commodity group are constant, Hicksian separability lets the
price of a single good represent the group price level. This is relied on by price questionnaires
used in household surveys. Methods of estimating demand systems from household survey data
also rely on Hicksian separability. Yet this restriction, and its weaker stochastic form under the
Generalized Composite Commodity Theorem, remains untested in cross-sections. We use unique
data from Vietnam with multiple specifications from within the same food groups to test if
within-group relative prices are constant over space. The data firmly reject these restrictions.