Frequently, the only available evidence in ‘language crimes’ (e.g. verbal sexual harassment) is witness statements about criminal conversations. However, previous research has showed that recall for sentences in conversation is very poor. The main aim of this research was to find out how to solve this problem. The cognitive interview (CI) is an interview technique which has shown to be more effective in recalling criminal episodes than a comparison interview. In addition, our experience in research on the CI had been highly satisfactory; therefore, we decided to use the CI as a tool in the research on memory for conversation, so far ignored. Thus, this study tested, for the first time, whether the CI would be also successful in obtaining complete and accurate accounts for a criminal conversation. Different forms of correct recall (verbatim/gist) of the verbal information as well as different types of errors (distortions/fabrications) were also examined. It was predicted that the CI would elicit more correct information without an increase in errors than a comparison interview (i.e. a free-recall protocol). Results confirmed these hypotheses. Interpretations of the overall findings are offered within the context of theoretical principles concerning the retrieval of information from memory. Copyright © 2008 John Wiley & Sons, Ltd.