Income-differentiated mortality, by reducing the share of poor persons in the population, leads to what can be called the "Mortality Paradox": the worse the survival conditions of the poor are, the lower the measured poverty is. We show that the extent to which FGT measures (Foster Greer Thorbecke 1984) underestimate old-age poverty under income-differentiated mortality depends on whether the prematurely dead would have, in case of survival, suffered from a more severe poverty than the average surviving population. Taking adjusted FGT measures with extended lifetime income profiles as a benchmark, we identify conditions under which the selection bias induced by income-differentiated mortality is higher for distribution-sensitive measures than for headcount measures. Finally, we show, on the basis of data on poverty in 11 European economies, that the size of the selection bias varies across different subclasses of FGT measures and across countries.