Background: Angiogenesis is a valid target in the treatment of epithelial ovarian cancer. Trebananib inhibits the binding of angiopoietins 1 and 2 to the Tie2 receptor, and thereby inhibits angiogenesis. We aimed to assess whether the addition of trebananib to single-agent weekly paclitaxel in patients with recurrent epithelial ovarian cancer improved progression-free survival. Methods: For this randomised, double-blind phase 3 study undertaken between Nov 10, 2010, and Nov 19, 2012, we enrolled women with recurrent epithelial ovarian cancer from 32 countries. Patient eligibility criteria included having been treated with three or fewer previous regimens, and a platinum-free interval of less than 12 months. We enrolled patients with a computerised interactive voice response system, and patients were randomly assigned using a permuted block method (block size of four) in a 1:1 ratio to receive weekly intravenous paclitaxel (80 mg/m2) plus either weekly masked intravenous placebo or trebananib (15 mg/kg). Patients were stratified on the basis of platinum-free interval (≥0 and ≤6 months vs >6 and ≤12 months), presence or absence of measurable disease, and region (North America, western Europe and Australia, or rest of world). The sponsor, investigators, site staff, and patients were masked to the treatment assignment. The primary endpoint was progression-free survival assessed in the intention-to-treat population. The trial is registered with ClinicalTrials.gov, NCT01204749, and is no longer accruing patients. Findings: 919 patients were enrolled, of whom 461 were randomly assigned to the trebananib group and 458 to the placebo group. Median progression-free survival was significantly longer in the trebananib group than in the placebo group (7·2 months [5·8-7·4] vs 5·4 months [95% CI 4·3-5·5], respectively, hazard ratio 0·66, 95% CI 0·57-0·77, p<0·0001). Incidence of grade 3 or higher adverse events was similar between treatment groups (244 [54%] of 452 patients in the placebo group vs 258 [56%] of 461 patients in the trebananib group). Trebananib was associated with more adverse event-related treatment discontinuations than was placebo (77 [17%] patients vs 27 [6%], respectively) and higher incidences of oedema (294 [64%] patients had any-grade oedema in the trebananib group vs 127 [28%] patients in the placebo group). Grade 3 or higher adverse events included ascites (34 [8%] in the placebo group vs 52 [11%] in the trebananib group), neutropenia (40 [9%] vs 26 [6%]), and abdominal pain (21 [5%] vs 22 [5%]). We recorded serious adverse events in 125 (28%) patients in the placebo group and 159 (34%) patients in the trebananib group. There was a difference of 2% or less in class-specific adverse events associated with anti-VEGF therapy (hypertension, proteinuria, wound-healing complications, thrombotic events, gastrointestinal perforations), except bleeding, which was more common in the placebo group than in the trebananib group (75 [17%] vs 46 [10%]). Interpretation: Inhibition of angiopoietins 1 and 2 with trebananib provided a clinically meaningful prolongation in progression-free survival. This non-VEGF anti-angiogenesis option for women with recurrent epithelial ovarian cancer should be investigated in other settings and in combination with additional agents. Although oedema was increased, typical anti-VEGF associated adverse events were not prominent. Funding: Amgen. © 2014 Elsevier Ltd.