Bank runs are a central concern for financial stability, yet systematic empirical evidence remains scarce. We construct a novel historical dataset of bank runs, covering 184 countries since 1800 by combining narrative evidence from 503 sources with statistical indicators of aggregate deposit contractions. We find that: (i) the unconditional likelihood of a bank run is 1.9%; (ii) systemic runs—those accompanied by aggregate deposit outflows—are associated with output losses of 9% over five years, more than after non-systemic runs or deposit contractions alone; (iii) these losses persist even when banks are well capitalized and there is no evidence of fundamental triggers, banking crises, or widespread bank failures; (iv) central banks and deposit insurance are linked to a lower probability of runs becoming systemic, while liability guarantees coincide with smaller output losses. Our findings highlight a key role of bank liability disruptions in economic fluctuations, over and above solvency issues.