This study examines the validity of a survey instrument on near-miss obstetric complications. Three groups of women--with severe complications, with mild complications, and with a normal delivery--were identified retrospectively in three hospitals in South Benin and interviewed at home. The concept of "near-miss" was used to identify women with severe episodes of morbidity. The questionnaire was able to detect, with some accuracy, eclamptic fits, abnormal bleeding in the third trimester for a recall period of at least three to four years, and all episodes of bleeding independent of timing within a period of two years. Questions concerning dystocia and infections of the genital tract generated disappointing results except when information on treatment was included. Overall, better results were achieved for antepartum and acute events. Severity made a positive difference only in the case of eclampsia, with an increase in sensitivity. The implications of the results for using women's recall of obstetric complications in surveys are discussed.