This comment speaks directly to the conceptual and methodological shortcomings of a single article (Szymanski, b), but has a broader purpose. Research reports concerning and/or employing income distributional measures have been appearing in the sociological literature with increasing frequency over the past few years. Many of these reports deal with the size distribution of subgroup income not for descriptive purposes, but to make important theoretical points. It is unfortunate that for far too long income distribution has been studied primarily by economists (when it has been studied at all, other than descriptively). This means that sociologists approaching the field without care run the considerable risk of unknowingly accepting the simplifying assumptions of many economists, some of whom are known as much for their concern with statistical rigor as for their seeming dismissal of questions of measurement validity.' Measurement invalidity is the bane of distributional research, and is probably the greatest problem currently facing students of the area.2 The greatest difficulty is that too many aren't even aware that it is a problem. Szymanski is but one case in point. I shall first review the problems of his analysis and interpretation, and then reanalyze his data to arrive at quite different conclusions. Szymanski investigates the economic impact on males of earnings discrimination against females. He concludes: . . while men as a whole do not earn more because of sexual discrimination . . . the poorer paid whites do benefit somewhat at the expense of women, while the better paid males lose. This conclusion follows, he claims, from two findings: (1) that male median income is unaffected by discrimination against females; and (2) that male inequality (as measured by the Gini Concentration Ratio) decreases as discrimination against females increases. Given these two, he says, males at the top must lose and those at the bottom must gain. His conclusions, of course, would not follow from his findings even if his findings were correct, which they are not. The assumption that they do follow is based on a misunderstanding of what the Gini is. The Gini is a notoriously poor measure of the well-being of any segment of a Lorenz curve (see Alker and Russet; Bronfenbrenner; Kravis).