Introducing a threshold in the sense of a minimal project size transforms a public-good game with an inefficient equilibrium into a coordination game with a set of Pareto-superior equilibria. Thresholds may therefore improve efficiency in the voluntary provision of public goods. In our one-shot experiment, we find that coordination often fails and exogenously imposed thresholds are ineffective at best and often counterproductive. This holds over a range of threshold levels and refund rates. We test whether thresholds perform better if they are endogenously chosen, i.e., whether a threshold is approved in a referendum, because voting may facilitate coordination due to signaling and commitment effects. We find that voting does have signaling and commitment effects, but they are not strong enough to significantly improve the efficiency of thresholds.