The National Football League (NFL) draft is used to examine the presence of the sunk-cost fallacy in teams’ playing time decisions. In the NFL, salary cap value represents a significant sunk cost to teams. We use the structure of the NFL draft to conduct a fuzzy regression discontinuity design. Optimal bandwidth local linear results suggest a 10% increase in salary cap value yields an additional 2.7 games started, for players selected near the cutoff between the first two rounds. Despite being no more productive, the first round selections receive a compensation premium, which leads to them starting significantly more games.