For most indexes in the Consumer Price Index (CPI), the Bureau of Labor Statistics (BLS) uses a form of the matched-model approach. It has long been recognized that the matched-model method can underestimate quality improvement, and therefore overestimate price inflation, for products exhibiting rapid technological improvement. In contrast, it is frequently taken for granted that the matched model approach accurately reflects price inflation for items that have no major trend in quality. In this paper we investigate that hypothesis using CPI data for retail food items. Our results provide evidence on the accuracy of the CPI, on the validity of models of retail firm pricing behavior, and on recent work on the stickiness of prices. The BLS handling of product replacements differs from a pure matched-model method. When an item disappears and is replaced in the CPI sample by another item, a CPI analyst in Washington decides whether the replacement product’s characteristics are comparable or non-comparable to those of the old product. If the two products are judged non-comparable, they are treated as wholly different models, and their qualityadjusted price difference is imputed from the price movements of other items in the sample. If the two products are judged comparable, however, their prices are used in the index without adjustment, in the same way as if no model change had taken place. As typically discussed, a matched-model index is one that imputes price change in all cases of product replacement. Our paper examines monthly price changes of continuing goods, comparable substitutes and non-comparable substitutes for a set of food items during a six-year period from 2003 through 2009. For substitutions, we compute both the actual price changes and estimated quality-adjusted changes. In previous work we selected 14 relatively homogeneous food categories in order to focus on price differences across stores. In this paper, we examine an expanded set of 39 categories, including many that comprise products with widely varying characteristics and more frequent turnover: for example, lunchmeats and snacks. We find, first, that price changes for comparable substitutions are, on average, sharply greater than price changes for continuing quotes. Second, comparing our quality-adjusted and unadjusted price changes for comparable substitutes, we cannot reject the hypothesis that they are equal on average. This indicates that the CPI analysts may be roughly correct in their decisions that no significant quality changes have taken place in those cases. It also supports the idea described most recently in Nakamura and Steinsson, (as opposed to Klenow and Willis) that firms take the opportunity of product replacement to make changes in the quality-adjusted price. Finally, for substitutes judged noncomparable, we find that on average the hedonically-adjusted price changes are significantly higher than the price changes imputed using the CPI method. This suggests that the current CPI procedure for non-comparable substitutions may underestimate price change, thereby overestimating quality change. Research on this issue should continue, and conclusions about the CPI treatment of product changes will, in turn, have potentially important implications for the measurement of real output and the response of prices to monetary policy.