Background Utidelone, a genetically engineered epothilone analogue, has shown promise as a potential treatment for breast cancer in phase 1 and 2 trials. The aim of this phase 3 trial was to compare the efficacy and safety of utidelone plus capecitabine versus capecitabine alone in patients with metastatic breast cancer. Methods We did a multicentre, open-label, superiority, phase 3, randomised controlled trial in 26 hospitals in China. Eligible participants were female patients with metastatic breast cancer refractory to anthracycline and taxane chemotherapy regimens. We randomly assigned participants (2: 1) using computer based randomisation and block sizes of 6 to a 21-day cycle of either utidelone (30 mg/m(2) intravenously once per day on days 1-5) plus capecitabine (1000 mg/m(2) orally twice per day on days 1-14), or capecitabine alone (1250 mg/m(2) orally twice per day on days 1-14), until disease progression or unacceptable toxicity occurred. Patients, physicians, and assessors were not masked to treatment allocation; however, an independent radiology review committee used to additionally assess response was masked to allocation. The primary endpoint was centrally assessed (by an independent radiology review committee) progression-free survival, and analysed using the Kaplan-Meier product-limit method in the intention-to-treat population. Safety was assessed in all participants who received at least one dose of study drug. Follow-up is ongoing. This study is registered at ClinicalTrials.gov, number NCT02253459. Findings Between Aug 8, 2014, and Dec 14, 2015, we enrolled and randomly assigned 270 patients to treatment with utidelone plus capecitabine, and 135 to capecitabine alone. Median follow-up for progression-free survival was 6.77 months (IQR 3.81-10.32) for the utidelone plus capecitabine group and 4.55 months (2.55-9.39) for the capecitabine alone group. Median progression-free survival by central review in the utidelone plus capecitabine group was 8.44 months (95% CI 7.95-9.92) compared with 4.27 months (3.22-5.68) in the capecitabine alone group; hazard ratio 0.46, 95% CI 0.36-0.59; p<0.0001. Peripheral neuropathy was the most common grade 3 adverse event in the utidelone plus capecitabine group (58 [22%] of 267 patients vs 1 [<1%] of 130 patients in the capecitabine alone group). Palmar-plantar erythrodysaesthesia was the most prominent grade 3 adverse event in the capacitabine alone group (in 10 [8%] of 130 patients) and was the next most frequent grade 3 event in the utidelone plus capecitabine group (in 18 [7%] of 267 patients). 16 serious adverse events were reported in the combination therapy group (diarrhoea was the most common, in three [1%] patients) and 14 serious adverse events were reported in the monotherapy group (the most common were diarrhoea, increased blood bilirubin, and anaemia, in two [2%] patients for each event). 155 patients died (99 in the combination therapy arm, 56 in the monotherapy arm). All deaths were related to disease progression except for one in each group (attributed to pericardial effusion in the combination therapy group and dyspnoea in the monotherapy group) that were considered possibly or probably treatment-related. Interpretation Despite disease progression with previous chemotherapies, utidelone plus capecitabine was more efficacious compared with capecitabine alone for the outcome of progression-free survival, with mild toxicity except for peripheral sensory neuropathy, which was manageable. The findings from this study support the use of utidelone plus capecitabine as an effective option for patients with metastatic breast cancer.