Objective To assess the accuracy of angiogenic biomarkers to predict pre-eclampsia. Design Prospective multicentre study. From 2006 to 2009, 5121 pregnant women with risk factors for pre-eclampsia (nulliparity, diabetes, previous pre-eclampsia, chronic hypertension) from Argentina, Colombia, Peru, India, Italy, Kenya, Switzerland and Thailand had their serum tested for sFlt-1, PlGF and sEng levels and their urine for PlGF levels at ≤20, 23-27 and 32-35 weeks' gestation (index tests, results blinded from carers). Women were monitored for signs of pre-eclampsia, diagnosed by systolic blood pressure ≥140 mmHg and/or diastolic blood pressure ≥90 mmHg, and proteinuria (protein/creatinine ratio ≥0.3, protein ≥1 g/l, or one dipstick measurement ≥2+) appearing after 20 weeks' gestation. Early pre-eclampsia was defined when these signs appeared ≤34 weeks' gestation. Main outcome measure Pre-eclampsia. Results Pre-eclampsia was diagnosed in 198 of 5121 women tested (3.9%) of whom 47 (0.9%) developed it early. The median maternal serum concentrations of index tests were significantly altered in women who subsequently developed pre-eclampsia than in those who did not. However, the area under receiver operating characteristics curve at ≤20 weeks' gestation were closer to 0.5 than to 1.0 for all biomarkers both for predicting any pre-eclampsia or at ≤34 weeks' gestation. The corresponding sensitivity, specificity and likelihood ratios were poor. Multivariable models combining sEng with clinical features slightly improved the prediction capability. Conclusions Angiogenic biomarkers in first half of pregnancy do not perform well enough in predicting the later development of pre-eclampsia.