The empirical literature on executive compensation generally fails to specify a model of executive pay on which to base and test hypotheses regarding its determinants. In contrast, this paper analyzes a simple principal-agent model to determine how well it explains variations in CEO incentive pay and salaries. Many findings are consistent with the basic intuition of principal-agent models that compensation is structured to trade off incentives with insurance. However, statistical significance for some of the effects is weak, although the magnitudes are large. Also, there is little evidence of the use of relative performance pay. Nevertheless, while puzzles remain, it seems clear that principal-agent considerations play a role in setting executive compensation.