Background The randomised phase 3 TURANDOT trial compared two approved bevacizumab-containing regimens for HER2-negative metastatic breast cancer in terms of efficacy, safety, and quality of life. The interim analysis did not confirm non-inferior overall survival (stratified hazard ratio [HR] 1·04; 97·5% repeated CI [RCI] –∞ to 1·69). Here we report final results of our study aiming to show non-inferior overall survival with first-line bevacizumab plus capecitabine versus bevacizumab plus paclitaxel for locally recurrent or metastatic breast cancer. Methods In this multinational, open-label, randomised phase 3 TURANDOT trial, patients aged 18 years or older who had an Eastern Cooperative Oncology Group performance status 0–2 and measurable or non-measurable HER2-negative locally recurrent or metastatic breast cancer who had received no previous chemotherapy for locally recurrent or metastatic breast cancer were stratified and randomly assigned (1:1) using permuted blocks of size six to either bevacizumab plus paclitaxel (bevacizumab 10 mg/kg on days 1 and 15 plus paclitaxel 90 mg/m2 on days 1, 8, and 15 every 4 weeks) or bevacizumab plus capecitabine (bevacizumab 15 mg/kg on day 1 plus capecitabine 1000 mg/m2 twice daily on days 1–14 every 3 weeks) until disease progression, unacceptable toxicity, or withdrawal of consent. Stratification factors were oestrogen or progesterone receptor status, country, and menopausal status. The primary objective was to show non-inferior overall survival with bevacizumab plus capecitabine versus bevacizumab plus paclitaxel in the per-protocol population by rejecting the null hypothesis of inferiority (HR ≥1·33) using a stratified Cox proportional hazard model. This trial is registered with ClinicalTrials.gov, number NCT00600340. Findings Between Sept 10, 2008, and Aug 30, 2010, 564 patients were randomised, representing the intent-to-treat population. The per-protocol population comprised 531 patients (266 in the bevacizumab plus paclitaxel group and 265 in the bevacizumab plus capecitabine group). At the final overall survival analysis after 183 deaths (69%) in 266 patients receiving bevacizumab plus paclitaxel and 201 (76%) in 265 receiving bevacizumab plus capecitabine in the per-protocol population, median overall survival was 30·2 months (95% CI 25·6–32·6 months) versus 26·1 months (22·3–29·0), respectively. The stratified HR was 1·02 (97·5% RCI –∞ to 1·26; repeated p=0·0070), indicating non-inferiority. The unstratified Cox model (HR 1·13 [97·5% RCI –∞ to 1·39]; repeated p=0·061) did not support the primary analysis. Intent-to-treat analyses were consistent with the per-protocol results. The most common grade 3 or worse adverse events were neutropenia (54 [19%] of 284 patients in the bevacizumab plus paclitaxel group vs 5 [2%] of 277 patients in the bevacizumab plus capecitabine group), hand–foot syndrome (1 [<1%] vs 43 [16%]), peripheral neuropathy (39 [14%] vs 1 [<1%]), leucopenia (20 [7%] vs 1 [<1%]), and hypertension (12 [4%] vs 16 [6%]). Serious adverse events were reported in 65 (23%) of 284 patients receiving bevacizumab plus paclitaxel and 68 (25%) of 277 receiving bevacizumab plus capecitabine. Deaths in two (1%) of 284 patients in the bevacizumab plus paclitaxel group were deemed by the investigator to be treatment-related. No treatment-related deaths occurred in the bevacizumab plus capecitabine group. Interpretation Bevacizumab plus capecitabine represents a valid first-line treatment option for HER2-negative locally recurrent or metastatic breast cancer, offering good tolerability without compromising overall survival compared with bevacizumab plus paclitaxel. Although progression-free survival with the bevacizumab plus capecitabine combination is inferior to that noted with bevacizumab plus paclitaxel, we suggest that physicians should consider possible predictive risk factors for overall survival, individual's treatment priorities, and the differing safety profiles. Funding Roche.