Background: Many trials of transcranial magnetic stimulation (TMS) have used small samples and, therefore, lack power. Here we present an up-to-date meta-analysis of TMS in the treatment of depression. Methods: We searched Medline and Embase from 1996 until 2008 for randomized sham-controlled trials, with patients and investigators blinded to treatment, and outcome measured using a version of the Hamilton Depression Rating Scale (or similar). We identified 1,789 studies. Thirty-one were suitable for inclusion, with a cumulative sample of 815 active and 716 sham TMS courses. Results: We found a moderately sized effect in favour of TMS [Random Effects Model Hedges’ g = 0.64, 95% confidence interval (95% CI) = 0.50–0.79]. The corresponding Pooled Peto Odds Ratio for treatment response (≤50% reduction in depression scores) was 4.1 (95% CI = 2.9–5.9). There was significant variability between study effect sizes. Meta-regressions with relevant study variables did not reveal any predictors of treatment efficacy. Nine studies included follow-up data with an average follow-up time of 4.3 weeks; there was no mean change in depression severity between the end of treatment and follow-up (Hedges’ g = –0.02, 95% CI = –0.22 to +0.18) and no heterogeneity in outcome. Discussion: TMS appears to be an effective treatment; however, at 4 weeks’ follow-up after TMS, there had been no further change in depression severity. Problems with finding a suitably blind and ineffective placebo condition may have confounded the published effect sizes. If the TMS effect is specific, only further large double-blind randomized controlled designs with systematic exploration of treatment and patient parameters will help to define optimum treatment indications and regimen.