Effectiveness of development aid is widely perceived to suffer in the presence of multiple donors with overlapping responsibilities. We test existing theory on aid fragmentation by studying aid provision under numerous donors throughout Afghanistan from 2006-2009. Our study leverages granular military data on aid and conflict, and household survey data on corruption and public opinion. We conduct the first micro-level analysis of aid fragmentation. When delivered by a single donor, aid appears to curtail corruption, boost public opinion, and reduce conflict. But under donor fragmentation, the benefits of aid are significantly reduced. Our results suggest under high volumes of aid provision, fragmentation facilitates corruption and thereby erodes aid’s ability to win hearts and minds in the fight against insurgents. At moderate levels of aid, however, fragmentation may actually benefit the quality of institutions. Our findings remain stable when accounting for a rich set of observable confounds. Moreover, we obtain robust estimates when correcting for bias likely arising from the omission of unobservable factors.