Objectives: Given the increasing prominence of cluster randomised trials, we sought to: (1) examine whether clustered trials produce baseline imbalances between intervention and control conditions; (2) examine whether individually randomised trials give results that are substantially different from cluster randomised controlled trails; (3) to explore the impact of methods that can be used to adjust for the inappropriate analysis of clustered trials. Methods: We used a dataset of 34 trials (12,294 participants) investigating the effectiveness of enhanced primary care for depression (‘collaborative care’). This included 14 cluster randomised trials and 20 individualised trials. We conducted a random effects meta-analysis to examine imbalance in baseline depression scores between intervention and control conditions. We used meta-regression to test for differential effect size at six months between clustered and individualised studies. Heterogeneity in relation to clustering was examined using the I² inconsistency statistic. Unit of analysis error was corrected using a range of plausible published ICCs of depression outcomes. Results: There were no baseline imbalances between intervention and control subjects in either clustered (p = 0.837) and unclustered (p = 0.737) studies. Clustered studies gave almost identical estimates of effect size when compared to individualised studies (SMDclustered = 0.25 95%CI 0.17 to 0.33; SMDindividualised = 0.24 95%CI 0.13 to 0.36) and were subject to less between study variability (I² = 68.6% vs. 0%). Adjustment for clustering within a range of estimates of intra-class correlation had minimal impact on pooled estimates, confidence intervals or overall clinical and statistical significance (pooled SMDICC 0.02 = 0.249 (95%CI 0.174 to 0.325) to SMDICC 0.05 = 0.258 (95%CI 0.172 to 0.345)). Conclusions: The additional effort and expense involved in clustered trials needs to be justified when individualised studies might produce robust and believable results. Techniques can be used to adjust for unit of analysis error when analysing the results of clustered studies, but the importance of clustering might have been exaggerated. More research is needed to examine the generalisability of these findings.