In a prospective cohort study of 267 children born to mothers infected with HIV-1 in New Haven, Connecticut, an abrupt decline in the risk of mother-to-child transmission occurred in 1990 and persisted at least through December, 1993. A retrospective, observational study was undertaken to identify factors that might be responsible for this decline. Three variables were assessed: the use of orally administered zidovudine during pregnancy, the CD4+ T-lymphocyte count of the mother, and the mode of delivery. The risk of transmission was 18.6% (36/194; 95% CI: 14.1-24.8%) in infants of all women not treated with zidovudine compared with 5.5% (3/55; 95% CI: 1.1-15.1%) in infants of all women who were treated (odds ratio: 0.25; p = 0.02). In a subgroup of women with known CD4+ cell counts, the risk of transmission was 21.1% (20/95; 95% CI: 13.4-30.6%) in untreated women compared with 5.5% (3/55) in those who received zidovudine (odds ratio: 0.22; p = 0.01). In women with CD4+ T-cell counts < 200/microl, the differences remained significant (39.1% in those not treated vs. 4.2% in those treated; p < 0.004). There was an inverse relationship between CD4+ cell count and risk of transmission: among untreated mothers whose T-lymphocyte counts were > or = 500, 200-499, or < 200/microl, HIV-1 was transmitted to the offspring of 8.2, 30.4, and 39.1% of offspring, respectively (p < 0.002 by the exact trend test). There was no significant association between mode of delivery and vertical transmission of HIV. We conclude that treatment with orally administered zidovudine alone (500 mg/day) in the course of routine prenatal care is associated with a significant reduction in the risk of vertical transmission.