Identifying when--during pregnancy, delivery or the postnatal period--transmission of human immunodeficiency virus (HIV) from mother to infant usually takes place is critical to the development of methods to prevent maternal-infant transmission. Evidence is reviewed in this paper as to whether transmission occurs prepartum (early or late in gestation), intrapartum, or postpartum with breast feeding. Evidence in support of the notion of prepartum transmission has come from isolation of HIV from aborted fetal organs, comparison of maternal-child viral genotypes and study of neonatal cell-mediated immune responses. Evidence against prepartum transmission is that fewer than half of the children later known to be HIV-infected can be identified by virological tests carried out close to birth. A reduced rate of transmission in infants delivered by Caesarean section, and a reduced risk of transmission to second-born twins delivered vaginally, offers support to the view that intrapartum factors influence the risk of HIV transmission. Transmission through breast feeding can occur if a mother is infected postpartum and seems to pose some additional risk if she is already infected at parturition. The risk of infection increases with the stage of maternal HIV disease, but specific immunological, clinical and viral characteristics need to be investigated further. A clinical trial of zidovudine, used during late pregnancy and delivery and given to the infant at birth, has reported a significant reduction in transmission. Primary prevention of HIV infection in women remains a principal priority.