To determine whether metronidazole reduces early preterm labour in asymptomatic women with positive vaginal fetal fibronectin (fFN) in the second trimester of pregnancy. Randomised placebo-controlled trial. Fourteen UK hospitals (three teaching). Pregnancies with at least one previous risk factor, including mid-trimester loss or preterm delivery, uterine abnormality, cervical surgery or cerclage. Nine hundred pregnancies were screened for fFN at 24 and 27 weeks of gestation. Positive cases were randomised to a week's course of oral metronidazole or placebo. Primary outcome was delivery before 30 weeks of gestation. Secondary outcomes included delivery before 37 weeks. The Trial Steering Committee (TSC) recommended the study be stopped early; 21% of women receiving metronidazole (11/53) delivered before 30 weeks compared with 11% (5/46) taking placebo [risk ratio 1.9, 95% confidence interval (CI) 0.72-5.09, P = 0.18]. There were significantly more preterm deliveries (before 37 weeks) in women treated with metronidazole 33/53 (62%) versus placebo 18/46 (39%), risk ratio 1.6, 95% CI 1.05-2.4. fFN was a good predictor of early preterm birth in these asymptomatic women; positive and negative predictive values (24 weeks of gestation) for delivery by 30 weeks were 26% and 99%, respectively (positive and negative likelihood ratios 15, 0.35). Metronidazole does not reduce early preterm birth in high risk pregnant women selected by history and a positive vaginal fFN test. Preterm delivery may be increased by metronidazole therapy.