We run a laboratory experiment to investigate how group size affects coordination in a bank-run game, in which participants choose simultaneously whether to withdraw or not and groups change over time. We find that bank-size significantly affects the withdrawal probability, which is on average 12% higher in large than in small banks. In the initial round(s), all groups exhibit a similar withdrawal rate of about 40%; then, large and medium banks converge to the bank-run equilibrium, while small banks exhibit a less clear-cut dynamic. In all banks, participants use past experience either to take ex-post rational decisions or to engage in experimentation, i.e., take in the current round the decision opposite to what was the best response in the previous round. We show that experimentation is a strategic choice of those subjects who attempt at promoting group convergence towards the efficient equilibrium.