Background: Depression is a common and important cause of morbidity and mortality worldwide. Depression is commonly treated with antidepressants and/or psychotherapy, but some people may prefer alternative approaches such as exercise. There are a number of theoretical reasons why exercise may improve depression. Objectives: To determine the effectiveness of exercise in the treatment of depression. Search strategy: We searched Medline, Embase, Sports Discus, PsycINFO, the Cochrane Controlled Trials Register, and the Cochrane Database of Systematic Reviews for eligible studies in March 2007. In addition, we hand-searched several relevant journals, contacted experts in the field, searched bibliographies of retrieved articles, and performed citation searches of identified studies. We also searchedwww.controlled-trials.com in May 2008. Selection criteria: Randomised controlled trials in which exercise was compared to standard treatment, no treatment or a placebo treatment in adults (aged 18 and over) with depression, as defined by trial authors. We excluded trials of post-natal depression. Data collection and analysis: We calculated effect sizes for each trial using Cohen's method and a standardised mean difference (SMD) for the overall pooled effect, using a random effects model. Where trials used a number of different tools to assess depression, we included the main outcome measure only in the meta-analysis. Main results: Twenty-eight trials fulfilled our inclusion criteria, of which 25 provided data for meta-analyses. Randomisation was adequately concealed in a minority of studies, most did not use intention to treat analyses and most used self-reported symptoms as outcome measures. For the 23 trials (907 participants) comparing exercise with no treatment or a control intervention, the pooled SMD was -0.82 (95% CI - 1.12, -0.51), indicating a large clinical effect. However, when we included only the three trials with adequate allocation concealment and intention to treat analysis and blinded outcome assessment, the pooled SMD was -0.42 (95% CI -0.88, 0.03) i.e. moderate, nonsignificant effect. The effect of exercise was not significantly different from that of cognitive therapy. There was insufficient data to determine risks and costs. Authors' conclusions: Exercise seems to improve depressive symptoms in people with a diagnosis of depression, but when only methodologically robust trials are included, the effect sizes are only moderate and not statistically significant. Further, more methodologically robust trials should be performed to obtain more accurate estimates of effect sizes, and to determine risks and costs. Further systematic reviews could be performed to investigate the effect of exercise in people with dysthymia who do not fulfil diagnostic criteria for depression.