This research investigated whether a witness's emotion could influence the accuracy of statements obtained by the use of a cognitive interview. A total of 70 first-year university students viewed a video depicting a road accident. Electrodes were attached to their arms in order to send fictitious electric shocks during the video (high-arousal condition) or to measure physiological signs (low-arousal condition). One week later, they were interviewed using either a cognitive interview (CI) or a structured interview (SI). It was hypothesised that the beneficial effect of the cognitive interview would be amplified by a high level of arousal, particularly concerning central aspects of the video. Results indicated that a CI elicited more correct central and peripheral details recalled, whatever the level of arousal inducted during the encoding of the to-be remembered event. Furthermore, high-arousal participants produced more accurate testimonies concerning peripheral details than participants exposed to a low level of arousal. No interaction between interview and emotion was found. The theoretical and practical implications for interviewing witnesses are described.