Several imputation methods have been developed for imputing missing responses. Often it is not clear which imputation method is "best" for a particular application. In choosing an imputation method, one should consider several factors, including the types of estimates that will be generated, the item nonresponse rates, the nature of the missing data, and the availability of auxiliary data that are correlated with the characteristic of interest or with the response propensity. This paper compares the performance of four commonly used imputation methods—nearest neighbor hot-deck, random hot-deck, cell mean, and regression imputation—in imputing missing benefit cost values for the Employment Cost Index (ECI) survey. It is assumed the nonrespondents are missing at random within classes. The advantages and disadvantages of each method are described; analytical and empirical results are presented.