Background: Meta-analysis of continuous outcomes typically uses mean difference (MD) or standardized MD (SMD) as effect measures. We recently described an alternative effect measure, the ratio of means (RoM), and demonstrated comparable statistical performance using comprehensive simulation. Others have suggested that the ratio of geometric means (RoGM) performs better for skewed data. SMD and ratio methods have the advantage of handling outcomes expressed in different units, but SMD has clinical limitations because the pooled standard deviation must be known. Objectives: To empirically compare treatment effects and heterogeneity for MD, SMD, RoM, and RoGM. Methods: The Cochrane Database of Systematic Reviews (2008, Issue 1) was searched for reviews pooling continuous data from >=5 trials. MD (if possible), SMD, RoM, RoGM were calculated for the outcome in each review with the most trials. Pairwise differences between methods for treatment effect and heterogeneity (using threshold p-values of 0.05 and 0.10 [Cochran’s Q], respectively) were assessed with Exact tests. Results: 232/5053 (4.6%) reviews combining 8 [5 to 10] (median [interquartile range]) trials met inclusion criteria. All methods demonstrated similar treatment effects. RoGM gave substantially higher heterogeneity. Heterogeneity for RoM was greater than SMD, possibly explained by the known negative bias of SMD decreasing its heterogeneity. Conclusions: Ratio and difference methods have similar significance of treatment effects, although RoGM has more heterogeneity. Empiric data and ease of interpretation justify use of RoM to report continuous outcomes in meta-analysis.