BACKGROUND: Associations between risk factors and perinatal outcomes may be biased at preterm gestational ages, if preterm delivery behaves as an effect modifier due to other unmeasured factors in the causal pathway. We evaluated whether fetuses-at-risk denominators could be used in regression models instead of conventional denominators to obtain less biased estimates of the association between maternal education and stillbirth at preterm gestational intervals. METHODS: Data included 2,143,134 live-born and 8946 stillborn singletons from 1981 through 2006 in Québec, Canada. Odds ratios and 95% confidence intervals were estimated for the relationship between education and stillbirth according to cause of fetal death, adjusting for maternal age, marital status, home language, parity, and period. We examined associations for 4 gestational intervals (<28, 28-31, 32-36, and ≥37 completed weeks), using both conventional denominators (ie, preterm live births) and fetuses-at-risk denominators. RESULTS: Stillbirth rates were greater for mothers with fewer years of education at all gestational intervals. Using conventional denominators, low education (relative to high education) was more strongly associated with term than preterm stillbirth and was apparently protective at <28 weeks. Using fetuses-at-risk denominators, low education was more strongly associated with preterm stillbirth than term stillbirth, even at <28 weeks. Low education was most strongly associated with diabetic-related stillbirth at ≥28 weeks (odds ratio = 5.04) relative to high education. CONCLUSIONS: Low education is associated with stillbirth throughout gestation, especially diabetic-related stillbirth. Use of fetuses-at-risk denominators in regression models can avoid potentially biased estimates obtained with conventional denominators at preterm gestational ages.