The bystander effect is the phenomenon that people are less likely to help others when they are in a group than when they are alone. The theoretical literature typically explains the bystander effect with the volunteer's dilemma: if providing help is equivalent to creating a public good, then bystanders could be less likely to help in groups because they free ride on the other bystanders. This paper uses a dynamic game to experimentally test such strategic interactions as an explanation for the bystander effect. In line with the predictions of the volunteer's dilemma, I find that bystanders help immediately when they are alone but help later and are less likely to help if they are part of a larger group. In contrast to the model's predictions, subjects in need of help are helped earlier and are more likely to be helped in larger groups. This finding can be accounted for in an extended model that includes both altruistic and selfish bystanders. The paper concludes that the volunteer's dilemma is a sensible way to model situations in which someone is in need of help, but it highlights the need to take heterogeneous social preferences into account.