Patient histories were obtained from 99 patients in three different ways: by a computerised patient interview (patient record), by the usual written interview (medical record), and by the transcribed record, which was a computerised version of the medical record. Patient complaints, diagnostic hypotheses, observer and record variations, and patients' and doctors' opinions were analysed for each record, and records were compared with the final diagnosis. About 40% of the data in the patient record were not present in the medical record. Two thirds of the patients said that they could express all or most of their complaints in the patient record. The doctors found that the medical record expressed the main complaints better (52%) than the patient record (15%) but that diagnostic hypotheses were more certain in the patient record (38%) than in the medical one (26%). The number of diagnostic hypotheses in the patient record was about 20% higher than that in the medical record. Intraobserver agreement (51%) was better than interobserver agreement (32%), while the inter-record agreement varied from 25% (between the medical and patient records) to 35% (between the transcribed and patient records). One third of final diagnoses were seen in the medical record, with 29% and 22% for the transcribed and patient records, respectively. Interobserver agreement in the final diagnosis was 35%. The results of the study suggest that computerised history taking is suitable for certain patients in addition to, and not as a substitute for, the oral interview with a doctor.