Sovereign debt crises are difficult to solve. This paper studies the “holdout problem”, meaning the risk that creditors refuse to participate in a debt restructuring. We document a large variation in holdout rates, based on a comprehensive new dataset of 23 bond restructurings with external creditors since 1994. We then study the determinants of holdouts and find that the size of creditor losses (haircuts) is among the best predictors at the bond level. In a restructuring, bonds with higher haircuts see higher holdout rates, and the same is true for small bonds and those issued under foreign law. Collective action clauses (CACs) are effective in reducing holdout risks. However, classic CACs, with bond-by-bond voting, are not sufficient to assure high participation rates. Only the strongest form of CACs, with single-limb aggregate voting, minimizes the holdout problem according to our simulations. The results help to inform theory as well as current policy initiatives on reforming sovereign bond markets.