The cognitive interview (CI) is a procedure designed for use in police interviews involving witnesses. This study tested the most recent version of the CI (Fisher & Geiselman, 1992) which comprises not only ‘cognitive’ techniques such as context reinstatement but also ‘social’ techniques for increasing rapport. Children (aged eight–nine years) viewed a magic show and were interviewed after a short delay of two days (time 1) and/or a longer delay of 12 days (time 2). At time 1 the CI produced a significantly greater amount of correct recall than did a structured interview (SI) which was similar to the CI save for the CI's special cognitive techniques. However, at time 1 the CI also produced significantly more recall errors. At time 2, no differences occurred between CI and SI recall. There was a significant hypermnesia effect at time 2 for those groups interviewed twice suggesting an effect of retrieval practice. The time 1 effects of the CI were found to exist only in the questioning phase of the interview and social and cognitive explanations for the changes in the nature of recall with a CI are considered. Practical implications are discussed in the context of good practice for interviewing child witnesses.