This paper offers an evaluation of a supported women's self-help program with over 1.5 million participants in one of the poorest rural regions of the world (Uttar Pradesh, India). Methodologically, it shows how indicators from the direct capability measurement literature can be adapted for program evaluation in a low-income country setting. Unique data on capabilities across a range of dimensions are then developed for some 6000 women and used to estimate a number of propensity score matching models. The substantive empirical results of these models indicate that many of the capability indicators are higher for program members, that the difference appears robust, and that there are significant benefits for those from scheduled tribes and lower castes. The discussion highlights two points. First, human development improvements offered by multi-strand programs can help to explain the paradox as to why nearly 100 million women (in India alone) have participated in self-help programs despite modest global research evidence for micro-finance impacts on nominal incomes. Second, results argue strongly for the use of capability measures over agency measures focused solely on household decision-making to assess women's empowerment when structural causes of disempowerment, external to the household, are present and significant.