Abstract. Background: Death certificates (DC) can provide valuable health status data regarding disease incidence, prevalence and mortality in a community. It can guide local health policy and help in setting priorities. Incomplete and inaccurate DC data, on the other hand, can significantly impair the precision of a national health information database. In this study we evaluated the accuracy of death certificates at a tertiary care teaching hospital in a Karachi, Pakistan. Methods. A retrospective study conducted at Aga Khan University Hospital, Karachi, Pakistan for a period of six months. Medical records and death certificates of all patients who died under adult medical service were studied. The demographic characteristics, administrative details, co-morbidities and cause of death from death certificates were collected using an approved standardized form. Accuracy of this information was validated using their medical records. Errors in the death certificates were classified into six categories, from 0 to 5 according to increasing severity; a grade 0 was assigned if no errors were identified, and 5, if an incorrect cause of death was attributed or placed in an improper sequence. Results: 223 deaths occurred during the study period. 9 certificates were not accessible and 12 patients had incomplete medical records. 202 certificates were finally analyzed. Most frequent errors pertaining to patients' demographics (92%) and cause/s of death (87%) were identified. 156 (77%) certificates had 3 or more errors and 124 (62%) certificates had a combination of errors that significantly changed the death certificate interpretation. Only 1% certificates were error free. Conclusion: A very high rate of errors was identified in death certificates completed at our academic institution. There is a pressing need for appropriate intervention/s to resolve this important issue.