In a real-effort experiment, we investigate how the timing of discretionary bonuses affects the relationship between workers and managers. Average output is substantially higher if bonuses are paid in the middle rather than upfront or at the end, as workers increase first-period output to signal trustworthiness. In contrast, average output does not differ when the decision is made at the beginning or end. When the decision is made upfront, output increases after receiving a bonus but decreases substantially, if the bonus is not paid. This is consistent with negative reciprocity by workers who anticipate, but do not receive a bonus.