A description is given of the system of replicated samples introduced into the Consumer Price Index as of December 1963. Essentially simple in concept, the main difficulties related to implementation of the design throughout the complex CPI structure. Estimates of sampling error in the CPI, the first such estimates for a major national price index, are given, and their adequacy and limitations evaluated. The results indicate that any change in the published CPI of .2 per cent (not index points) over a monthly, quarterly or longer period is significant at the 5 per cent level.