We have examined the accuracy of text entries within a manually compiled anaesthetic record by comparing the record of the anaesthetist with that of an observer, present throughout the procedure but whose sole purpose was the documentation of perioperative events. Eighty-six items of information were analysed for accuracy from 197 records. The mean proportion of omissions was 35% and the mean proportion of incorrect entries was 3.4%. Where no entry should have been made, the mean proportion of unwarranted entries was 1%. Accuracy varied according to the information contained; however, omissions were common for preoperative status, fluids, tourniquet use, aspects of monitoring, local anaesthesia and intraoperative problems. The most consistently accurate information was the description of the patient and that relating to intraoperative i.v. drug use. Inaccuracy was common for the majority of sites on the record, irrespective of their reflecting on the anaesthetist's performance. We suggest, therefore, that the reason for this inaccuracy of data was not related principally to anaesthetists' defensiveness, but may reflect their attitudes to the record's value and response to inadequacies in its design. The observed deficiencies in recording accuracy may affect patient safety during future anaesthesia and has relevance to medico-legal and epidemiological research.