To examine whether risk factors differed among subgroups of preterm (< 37 weeks of gestation) deliveries, we studied a cohort of 1,825 enlisted servicewomen who delivered from 1987 through 1990 at four U.S. Army medical centers. Preterm deliveries were classified by length of gestation (< 29 weeks, 29-32 weeks, 33-36 weeks) and clinical course [medical indication, idiopathic preterm labor, or preterm rupture of membranes (PROM)]. We abstracted medical records for information on age, race, army rank, marital status, gravidity, parity, the baby's sex, maternal prepregnancy height and weight, gestation at entry to prenatal care, alcohol drinking and smoking, time since and outcome of preceding pregnancy, surgery performed during pregnancy, anemia, and diagnoses of uterine abnormalities, sexually transmitted diseases, and urinary tract infections. We used proportional hazards analysis to evaluate associations for each subgroup of preterm delivery. The relative odds associated with a history of preterm delivery in the preceding pregnancy ranged from 3.1 for deliveries due to preterm labor or PROM to 6.2 for deliveries that occurred during 29-32 weeks; none of the other factors was consistently associated across the subgroups of preterm delivery. The paucity of associations is consistent with the conclusion of other investigators that most of the causes of preterm delivery are unknown.