Background: Multiple treatment meta-analysis (MTM) has become increasingly popular over the past years, but the area is still underresearched. Analogous to the importance of dealing with discrepancies that lie in between trials (heterogeneity) in pair wise comparison metaanalysis, MTM also encompasses the challenge of dealing with discrepancies that lie in between comparisons (inconsistency) in the network of head-to-head comparisons. Inconsistency should be reduced to facilitate more reliable synthesis of treatment effects. Study-specific covariates may account for the inconsistency; previous studies have accomplished reduction of inconsistency by incorporating a few covariates in the multivariate logistic regression models that facilitate MTM. However, in most systematic reviews where MTM is desirable the included studies rarely report all study-specific variables, and moreover, usually report different selections of these covariates. The problem is then one of missing data. Presumably, if full data sets are made available through imputation, model selection procedures can be utilised to facilitate reduction of inconsistency. Objectives: To demonstrate how multiple imputation of missing covariates may facilitate model selection in MTM for reducing inconsistency. Methods: We considered data from the Cochrane systematic review ‘‘Drugs for preventing postoperative nausea and vomiting’’ which included 737 studies investigating 46 different treatments. We limited our analysis to the six most frequently examined treatments. This subset included 226 trials and head-to-head comparison between all treatments. For each study we had recorded 18 covariates. The proportion of missing covariates was approximately 15%. We ran the full regression model using WinBUGS and applied multiple imputation by imputing from the posterior distributions of the missing data cells. We selected the best model fit based on the deviance information criterion (DIC). Finally, we compared the inconsistency in the best-fit model to the standard MTM model precluding covariates. Results: The best-fit model based on DIC showed an apparent reduction in inconsistency compared to the MTM model precluding covariates. The results will be presented in detail at the Colloquium. Conclusions: Multiple imputation in combination with model selection may provide a novel apprach for reducing inconsistency in multiple treatment meta-analysis.