The distinct value of separation of powers in democracy is its effect on electoral accountability. Among others, this paper focuses on how separation of powers controls the moral hazard of agents. The key feature of separation of powers in policymaking is that the results of rejected policies are not observed but inferred by the voter and that the voter’s inference is crucial in reelection decisions. In this setting, how and when the agents behave in the voter’s interest are far from straightforward. This paper shows the mechanism of separation of powers and argues that separation of powers works when 1) reelection is important enough for the agents compared to policy outcomes or 2) the veto player’s interest is not closely aligned with the proposer’s, and that it works best when both of the conditions are met. These results not only support the intuition that separation of powers works better when the interests of agents conflict with each other but also show that the two interconnected dimensions matter for separation of powers to work.