Consumers are very responsive to sales, yet statistical agency practice typically under-weights sale prices in the Consumer Price Index (CPI). Evidence is lacking on the impact on the representativeness of prices included in the CPI and on estimates of inflation. Using high-frequency scanner data from US supermarkets we find that the exclusion of sales prices introduces a systematic but stable effect over time so inflation measurement is not significantly affected. We also find evidence that the typical practice of using data from an incomplete period in constructing unit values can lead to an upward bias in the resulting price index.