OBJECTIVE To determine whether maternal serum levels of alphafetoprotein (alpha-FP) and human chorionic gonadotrophin (hCG) at 15-21 weeks provided clinically useful prediction of stillbirth in first pregnancies. METHODS Retrospective study of record linkage of a regional serum screening laboratory to national registries of pregnancy outcome and perinatal death. METHODS West of Scotland, 1992-2001. METHODS A total of 84,769 eligible primigravid women delivering an infant at or beyond 24 weeks of gestation. METHODS The risk of stillbirth between 24 and 43 weeks was assessed using the Cox proportional hazards model. Logistic regression models within gestational windows were then used to estimate predicted probability. Screening performance was assessed as area under the receiver operating characteristic (ROC) curve. METHODS Antepartum stillbirth unrelated to congenital abnormality. RESULTS The odds ratio (95% CI) for stillbirth at 24-28 weeks for women in the top 1% were 11.97 (5.34-26.83) for alpha-FP and 5.80 (2.19-15.40) for hCG. The corresponding odds ratios for stillbirth at or after 37 weeks were 2.44 (0.74-8.10) and 0.79 (0.11-5.86), respectively. Adding biochemical to maternal data increased the area under the ROC curve from 0.66 to 0.75 for stillbirth between 24 and 28 weeks but only increased it from 0.64 to 0.65 for stillbirth at term and post-term. Women in the top 5% of predicted risk had a positive likelihood ratio of 7.8 at 24-28 weeks, 3.7 at 29-32 weeks, 5.1 at 33-36 weeks and 3.4 at 37-43 weeks, and the corresponding positive predictive values were 0.97, 0.33, 0.47 and 0.63%, respectively. CONCLUSIONS Maternal serum levels of alpha-FP and hCG were statistically associated with stillbirth risk. However, the predictive ability was generally poor except for losses at extreme preterm gestations, where prevention may be difficult and interventions have the potential to cause significant harm.