Does motivated reasoning harm democratic accountability? Substantial evidence from political behavior research indicates that voters have "directional motives" beyond accuracy, which is often taken as evidence that they are ill-equipped to hold politicians accountable. We develop a model of electoral accountability with voters as motivated reasoners. Directional motives have two effects: (1) divergence -- voters with different preferences hold different beliefs, and (2) desensitization -- the relationship between incumbent performance and voter beliefs is weakened. While motivated reasoning does harm accountability, this is generally driven by desensitized voters rather than polarized partisans with politically motivated divergent beliefs. We also analyze the relationship between government performance and vote shares, showing that while motivated reasoning always weakens this relationship we cannot infer that accountability is also harmed. Finally, we show that our model can be mapped to standard models in which voters are fully Bayesian but have different preferences or information.