OBJECTIVE: To estimate the association between birth weight percentile and the risk of perinatal death at term in relation to the cause of death. METHODS: We performed a retrospective cohort study of all term singleton births in delivery units in Scotland between 1992 and 2008 (n=784,576), excluding perinatal deaths ascribed to congenital anomaly. RESULTS: There were 1,700 perinatal deaths in the cohort, which were not the result of congenital anomaly (21.7/10,000 women at term). We observed a reversed J-shaped association between birth weight percentile and the risk of antepartum stillbirth in all women, but the associations significantly differed (P<.001) according to smoking status. The highest risk (adjusted odds ratio referent to 21st-80th percentile, 95% confidence interval) among nonsmokers was for birth weight third or less percentile (10.5, 8.2-13.3), but there were also positive associations for birth weight percentiles 4th-10th (3.8, 3.0-4.8), 11th-20th (1.9, 1.5-2.4), and 98th-100th (1.8, 1.3-2.4). Among smokers, the associations with being small were weaker and the associations with being large were stronger. We also observed a reversed J-shaped association between birth weight percentile and the risk of delivery-related perinatal death (ie, intrapartum stillbirth or neonatal death), but there was no interaction with smoking. The highest risk was for birth weight greater than the 97th percentile (2.3, 1.6-3.3), but there were also associations with third or less percentile (2.1, 1.4-3.1), 4th-10th (1.8, 1.4-2.4), and 11th-20th (1.5, 1.2-2.0). Analysis of the attributable fraction indicated that approximately one in three antepartum stillbirths and one in six delivery-related deaths at term could be related to birth weight percentile outside the range 21st-97th percentile. CONCLUSION: Effective detection of variation in fetal size at term has potential as a screening test for the risk of perinatal death. LEVEL OF EVIDENCE: II.