This paper focuses on the targeting accuracy of NREG in two Indian states, Madhya Pradesh (MP) and Tamil Nadu (TN), based on household data for 2008–09. In order to overcome the difficulties arising from the use of a headcount index and a specific poverty threshold, stochastic dominance tests are used. Madhya Pradesh demonstrated much better targeting than Tamil Nadu in terms of the FGT class of poverty indices over a wide range of poverty thresholds. This finding is significant as the proportion of poor is twice as high in the former as in the latter. It raises doubts about prevailing views that there is greater underprovision of jobs under NREG in the poorer states. That the self-selection of the poor was undermined and (relatively) affluent crowded in because of the high NREG wage (relative to the agricultural wage) raises a serious concern. Another serious concern is that the transfer benefits in the form of additional income to the poor were small mainly due to short spells of work, considerably lower than the maximum number of days permitted under this scheme.