Abstract This paper develops a method to decompose differences across distributions of
household income, based on counterfactual distributions that ‘lie between’ the actually
observed distributions. Our approach decomposes differences between any two income
distributions (or functionals such as inequality or poverty measures) into shares due to price
effects; occupational structure effects; and endowment effects. Comparing the household
income distributions of the USA and Brazil in 1999, we find that most of Brazil’s excess
inequality (of 13 Gini points) is accounted for by underlying inequalities in the distributions
of education and of non-labor income, notably pensions (between four and six Gini points
each). Steeper returns to education in Brazil also make an important contribution (of two to
five points). Differences in occupational structure and in racial and demographic
composition are much less important.