The validity of registered official cause of death (COD) concerning breast cancer patients in Malmö 1964-1992 was analysed. There was a high frequency of autopsy in the material and, therefore, the number of diagnostic errors should be low. Found discordance thus mostly reflect the quality of certifying and registration of COD. Of 2631 deaths discordance was found as to underlying COD in 121 cases (4.6%). Ninety-four of these represent outright errors. In 27 there was doubt about the reported COD. The number of discordant cases increased with age at death. Two kinds of systematic bias were found. One was a tendency to report breast cancer as contributing COD in patients with clinical cure. The other was a practice at the Registry to register breast cancer as underlying COD in cases with breast cancer reported as contributing COD. This practice was changed in 1981 causing an artificial decrease in registered breast cancer mortality.