Households around the world believe inflation is much higher than observed in official statistics. I propose a model of expectation formation consistent with the apparent disconnect between beliefs about inflation and reality. In the model, households acquire information about prices while shopping and use this information to form beliefs about current and future aggregate inflation. Obtaining information is costly, and shoppers choose it optimally to trade it’s costs and benefits. The model makes predictions relating household’s characteristics with their shopping experiences and their beliefs about aggregate inflation. I test these predictions using microdata from the New York Fed’s Survey of Consumer Expectations and find evidence supporting the model’s predictions. First, households’ expectations about aggregate inflation respond systematically to their expectations about individual prices, and the strength of this response is proportional to the expenditure share of that good. Second, individual forecasts about prices are predicted by past forecasts after controlling extensively for sources of public information. Third, households that spend more on goods like gasoline are relatively better forecasting the price of those goods. To study the importance of inattention to prices for macroeconomic policy, I extend the standard New Keynesian model to include rationally inattentive shoppers. Information frictions produce a potentially large wedge between CPI inflation and shoppers’ perceived inflation. Despite this disagreement, a central bank targeting CPI inflation can stabilize the economy by following the Taylor principle. However, if information frictions are too large, responding aggressively to CPI inflation leads the economy to indeterminacy. The results suggest that strict inflation targeting can be dangerous in a world where household’s form beliefs about inflation based on shopping experiences.