This paper examines the extent to which gains-from-trade predictions from commonly-used trade theories are consistent with observed household consumption decisions. Our approach is based on inference from household-level estimation of food Engel curves in the US and in a few other countries. For a given price index as the deflator of income, deviations from food Engel curves indicate how biased that price index is relative to the true household price index. We construct open-economy price indices based on trade theory and data, evaluate their biases according to our approach, and compare them with the bias of official CPI statistics. We find that theory-consistent open-economy price indices that account for industry-level heterogeneity and input-output linkages tend to eliminate a large fraction of the bias of CPI.