We investigate the backdating of stock option exercises. Before SOX, we find evidence that some exercises were backdated to days with low stock prices. Consistent with a tax-based incentive, these suspect exercises are more likely when the personal tax savings from backdating are higher. However, suspect CEO exercises generate average (median) estimated tax savings of 96,000 (7,000). These savings appear modest relative to the costs insiders and firms face. We find that the likelihood of a suspect exercise increases in the likelihood of option grant backdating. This suggests that agency problems associated with backdating permeate option compensation in some firms.