Several tests of targeting accuracy of the National Rural Employment Guarantee Scheme (NREG) focusing on shares of participants by poverty status, their duration of participation, and earnings from it are used. The analysis is based on primary household data collected from three India states, Rajasthan, Andhra Pradesh and Maharashtra. In all three states, the poor depended more on the NREG than the non-poor, with the share of NREGS earnings in household income of the poor being the highest in Andhra Pradesh. Useful insights into the design and implementation of this scheme that impede the participation of the poor and render it more attractive for the (relatively) affluent are obtained from a probit analysis. A major flaw is the hike in the NREG wage relative to agricultural wage, as it undermines selfselection of the poor-especially in villages with a high degree of land inequality. In fact, two different mechanisms seem to be operating-one tends to exclude the poorest (the negative effect of the land Gini), and the other tends to promote the inclusion of the (relatively) affluent (the positive effect of the interaction of the land Gini and the ratio of NREG wage to agricultural wage). That awareness of the scheme matters is corroborated. However, the poor do not necessarily benefit as much as the non-poor at the entry point. But, with more information, corruption reduces at the implementation stage and this has the potential of serving the interests of the poor better.