Flows of official development assistance (ODA) to recipient countries have been highly volatile over the past 40 years, and there is significant evidence that volatile aid can negatively impact growth through several channels; but less is known about the sources of the volatility. Using an auto-regressive conditional heteroskedasticity model, we generate conditional variances for total aid flows to all aid-recipient countries between 1960 and 2008. We then examine the effects of both recipient-country and donor-related factors on this resulting volatility. We find that some degree of volatility is caused by events in recipient countries, mainly civil wars and adverse regime change—all of which increase the unpredictability of aid flows. But larger, unexpected swings in aid tend to be due to the concentration of aid portfolios combined with the prevalence of donor herding. Our results demonstrate, additionally, that the United States is the most volatile aid-giver, but that volatility is mostly due to U.S. aid recipients receiving unanticipated aid windfalls. Our findings are consistent when we remove aid flows for humanitarian assistance, emergency relief, food aid, technical assistance and debt relief. These results demonstrate the need for donor action in mitigating aid volatility.