We searched MEDLINE, EMBASE and the Cochrane Controlled Trials Register to determine whether oxygen relieves dyspnoea in mildly or non-hypoxemic COPD and included 18 randomised controlled trials (431 participants) in the meta-analysis using Cochrane methodology. Oxygen therapy reduced dyspnoea when compared with medical air; standardised mean difference -0.37 (95% CI -0.50 to -0.24; I(2)=14%). In a priori subgroup and sensitivity analyses, dyspnoea was reduced by continuous oxygen during exertion but not short-burst oxygen therapy. Continuous exertional oxygen can relieve dyspnoea in mildly or non-hypoxemic COPD, but evidence from larger clinical trials is needed.