The article presents a methodology for comparing the search and retrieval functions used by two different categories of translation memory tool. First the underlying differences between tools that use a sentence-based approach and those that use a character-string-within-a-bitext (CSB)-based approach are discussed. It is then explained how the 'usefulness" of the information that each tool retrieves can be defined and measured using a series of scores that take into account the validity of the information retrieved as well as the time that is gained or lost by using the tool. The results of a practical experiment conducted using representative tools from each category are presented. The article concludes with some observations about the types of text and user that best suited to each of the two types of translation memory that are discussed.