In this paper we describe a method of assessing eyewitness performance for actual crimes that could prove a valuable addition to more traditional experimental and field-based approaches to the study of eyewitnessing. We present the findings of the first reported attempt to assess the accuracy of information contained in police statements given by eyewitnesses to actual criminal episodes using CCTV as a means of verification. Employing the criterion of using only those items that could be verified against CCTV recordings (largely action details), both victims and bystanders from eight incidents of assault were found to be highly accurate in their accounts (96% accurate). These results are discussed in terms of what they might indicate about the relationship between arousal and eyewitness performance and how they compare with laboratory and other field-based approaches to the study of eyewitness memory. In addition, we consider some of the methodological, technological and practical constraints associated with this novel approach and its possible future applications to the study of everyday memory as well as memory for unusual events.