This paper shows that accounting for variation in mistakes can be crucial for welfare analysis. Focusing on consumer underreaction to not-fully-salient sales taxes, we show theoretically that the efficiency costs of taxation are amplified by 1) individual differences in under reaction and 2) the degree to which attention is increasing with the size of the tax rate. To empirically assess the importance of these issues, we implement an online shopping experiment in which 2,998 consumers-matching the U.S. adult population on key demographics-purchase common household products, facing tax rates that vary in size and salience. We find that: 1) there are significant individual differences in underreaction to taxes. Accounting for this heterogeneity increases the efficiency cost of taxation estimates by at least 200%, as compared to estimates generated from a representative agent model. 2) Tripling existing sales tax rates roughly doubles consumers' attention to taxes. Our results provide new insights into the mechanisms and determinants of boundedly rational processing of not-fully-salient incentives, and our general approach provides a framework for robust behavioral welfare analysis.