Many studies have examined associations between sociodemographic variables and preterm birth in singletons. However, almost no research has been published on whether variables such as maternal age, race, ethnicity, level of education and smoking are associated with preterm birth among twins in the same way. The purpose of this study was to examine such associations in twins and singletons comparatively. The study population consisted of all 567796 twins and 23297909 singleton births recorded in the US birth records for 1990-95. Gestational age data were rigorously 'cleaned' to solve the problem of biologically implausible birthweight/gestation combinations in vital records. Preterm birth was defined as gestational age < 35 weeks. Some 25.8% of twins and 3.2% of singletons were preterm by this definition. Crude and adjusted relative risks (RR) were estimated using a modified Mantel-Haenszel procedure. We found several characteristics associated with preterm birth in both twins and singletons, e.g. for twins: race (black adjusted RR = 1.30 compared with white non-Hispanic); marital status (unmarried adjusted RR = 1.15 compared with married); and age (< or = 17 years adjusted RR = 1.39 compared with 20-29 years). A similar analysis of singletons revealed stronger associations between the same characteristics and preterm birth, e.g. the adjusted RR for black race was 2.3. These differences in RRs suggest that sociodemographic characteristics have weaker effects on preterm birth among twins than among singletons. Care must be taken in interpreting differences in preterm birth in twins and singletons, as their gestational age distributions differ so markedly.