Although maternal mortality is higher in France than in other European countries, hospital data prove that maternal deaths are underestimated. To assess the degree of underestimation and investigate the reasons for it, a retrospective survey was carried out among the certifying doctors of the 3045 deaths that occurred among women aged 15-44 years, from December 1988 to March 1989. Doctors were asked for information on the obstetric condition of the women and their health before death. Some 88.2% of those approached responded. Sociodemographic information was obtained from the French national record of causes of death. Although doctors reported gravid puerperal conditions on 41 death certificates, only 24 deaths were classified in the maternal mortality category of the International Classification of Diseases. The other 17 deaths were classified elsewhere. The present survey permitted the identification of 27 additional deaths of which 16 were considered as having obstetric causes. Of the 68 deaths in pregnant or puerperal women which occurred during the survey period, 54 were classified as having obstetric causes. No sociodemographic differences were found between the deaths registered in the national record and the newly identified maternal deaths. All deaths occurring during or after parturition were reported to the national record, but most of the deaths from abortion were identified from the survey. The discussion deals with the misclassification of maternal deaths and the difficulty of determining the underlying cause of deaths involving complex diseases or uncertain pathogenesis.