BACKGROUND: During a retrospective study of intra-abdominal infections we found that case notes for adult patients were destroyed or incompletely microfiched if the patient had no contact with the hospital for six years. OBJECTIVES: To test the hypothesis that patients with unobtainable case notes would have higher in-hospital mortality and that this would bias the study sample. METHODS: Patients who might have community acquired intra-abdominal infections were identified at three acute hospitals from Scottish Morbidity Records (SMR) for 1993 to 1995 by the following ICD9 codes: 567 (peritonitis); 540 (appendicitis); 531-534 (gastric, duodenal, peptic or gastrojejunal ulcers with perforation). Medical records were obtained to validate the diagnosis and to obtain details of antibiotic therapy and its outcome. If medical records were unobtainable the date of death and duration of admission were identified from the electronic SMR record. RESULTS: A total of 867 cases were identified. For 339 (39%) of these, case notes were unobtainable or incomplete. SMR could be used to calculate in hospital mortality and duration of stay for 334 (98%) of patients with unobtainable or incomplete medical records and 510 (97%) of patients with complete notes. The proportion of female patients was similar (51% versus 50%). However, the patients with unobtainable or incomplete records were significantly older (median age 72 versus 53.5; p=0.0001 Wilcoxon test) and significantly more of them died in hospital (45% versus 7%; relative risk of death in hospital 6.75; 95% CI 5.08 to 8.96). Median length of stay (10 days versus 9 days; p=0.27 Wilcoxon test) and total hospital cost ( 2885 versus 2746; p=0.68) were similar in the two groups. CONCLUSION: Missing data have biased the study sample by excluding older patients with worse outcome. Selective destruction of case notes is an important source of bias in retrospective outcomes studies.