We study the impact of fossil fuel subsidy removal on presidential popularity using difference-indifference approaches and a stylized theoretical model. Analyzing macro level data for two subsidy removal events in Mexico and Bolivia in the early 2010s, we find evidence of a negative impact on presidential approval. Our theoretical probabilistic voting model predicts that the decline in popularity is driven by high income groups if subsidies are regressive, and that lack of trust in the government lowers popularity of the removal in all income groups. We confirm these predictions using micro level data for the Mexican subsidy removal event.