Angus Deaton discusses the ambiguity that arises from using different definitions and data sources for individual income or consumption levels in world poverty measurement. Should one rely on the direct information on individual consumption or income provided by national representative household surveys, or should consumption and income figures be scaled up or down so that means coincide with National Accounts (NA) data? It is generally the case that consumption expenditure per capita estimated in the NA is higher than the mean expenditure per capita obtained in surveys: thus scaling up leads to lower poverty estimates than when surveys are used. It is also the case that the difference between the two estimates tends to widen over time, so that trends are not more reliable than poverty estimates at one point of time. Deaton analyzes in detail the reasons for this divergence and concludes that NA-scaled survey data are in some sense faulty, whereas a pure consistency argument pleads in favor of using survey data at their face value.