We study an important recent series of multi-item multi-unit auctions conducted by the U.S. Treasury in retiring $67.5 billion of its debt. Consistent with auction theory, we find that bidders earn a small volatility-related expected profit as compensation for bearing the risk of the â€œwinnerâ€™s curse.â€ We find that the Treasury is penalized for being â€œspread too thinâ€ when including multiple bonds in a buyback auction. Thus, the multi-item design of the auction may not have been optimal from the Treasuryâ€™s perspective. In contrast, there is no evidence that the multi-unit aspect of the buyback affected the Treasuryâ€™s costs. These results have a number of implications for current models of multi-item and multi-unit auctions.