Background: Non-randomised trials have reported benefits of kyphoplasty in patients with cancer and vertebral compression fractures (VCFs). We aimed to assess the efficacy and safety of balloon kyphoplasty compared with non-surgical management for patients with cancer who have painful VCFs. Methods: The Cancer Patient Fracture Evaluation (CAFE) study was a randomised controlled trial at 22 sites in Europe, the USA, Canada, and Australia. We enrolled patients aged at least 21 years who had cancer and one to three painful VCFs. Patients were randomly assigned by a computer-generated minimisation randomisation algorithm to kyphoplasty or non-surgical management (control group). Investigators and patients were not masked to treatment allocation. The primary endpoint was back-specific functional status measured by the Roland-Morris disability questionnaire (RDQ) score at 1 month. Outcomes at 1 month were analysed by modified intention to treat, including all patients with data available at baseline and at 1 month follow-up. Patients in the control group were allowed to crossover to receive kyphoplasty after 1 month. This study is registered with ClinicalTrials.gov, NCT00211237. Findings: Between May 16, 2005, and March 11, 2008, 134 patients were enrolled and randomly assigned to kyphoplasty (n=70) or non-surgical management (n=64). 65 patients in the kyphoplasty group and 52 in the control group had data available at 1 month. The mean RDQ score in the kyphoplasty group changed from 17·6 at baseline to 9·1 at 1 month (mean change -8·3 points, 95% CI -6·4 to -10·2; p<0·0001). The mean score in the control group changed from 18·2 to 18·0 (mean change 0·1 points; 95% CI -0·8 to 1·0; p=0·83). At 1 month, the kyphoplasty treatment effect for RDQ was -8·4 points (95% CI -7·6 to -9·2; p<0·0001). The most common adverse events within the first month were back pain (four of 70 in the kyphoplasty group and five of 64 in the control group) and symptomatic vertebral fracture (two and three, respectively). One patient in the kyphoplasty group had an intraoperative non-Q-wave myocardial infarction, which resolved and was attributed to anaesthesia. Another patient in this group had a new VCF, which was thought to be device related. Interpretation: For painful VCFs in patients with cancer, kyphoplasty is an effective and safe treatment that rapidly reduces pain and improves function. Funding: Medtronic Spine LLC.