This study compared the realism of witnesses' confidence judgments of their own recall elicited in either the Cognitive Interview (CI) or the Structured Interview (SI). Calibration methodology was used to analyse the realism of the confidence judgments. Participants ( N =56) were interviewed about their observations of a staged crime and returned 2 weeks later to assess their confidence in the statements. The CI, as in previous research, rendered a significantly higher number of correct recalled units compared with the SI. In both interview conditions the witnesses displayed a high level of accuracy and confidence. In addition, a very good level of calibration and very low overconfidence was found in the witnesses' confidence judgments. No difference was found between the CI and SI in terms of the realism of the witnesses' confidence judgments, as measured by level of calibration or under/overconfidence. Furthermore, the CI and the SI did not differ with respect to the witnesses' estimations of the total number of correctly recalled items. It is suggested that the good realism found in these measures is due to the fact that witnesses assessed their confidence in items they had chosen to report of their own accord.