OBJECTIVES Cost-effectiveness models are an important component of health economic evaluation. In addition to estimates of treatment effects (typically estimated for RCTs), cost-effectiveness estimates may be sensitive to estimates of non-treatment specific parameters that describe the relationship between model variables. These may be estimated from epidemiological studies that themselves include covariable adjustment in order to account for potential confounding. We use the example of the estimated relationship between BMI and utility as to illustrate methods for the review meta-analysis of parameter estimates arising from multiple studies and issues around the selection of appropriate estimates. METHODS A targeted search was carried out in MEDLINE and EMBASE for studies with utility data on BMI. The outcome was utility change per unit increase in BMI. Study characteristics recorded included the utility instrument used, study location, diabetes status and number of covariates. Fixed and random effects models as well as graphical methods were used to investigate the influence of study characteristics. RESULTS Several utility scales were used throughout with some using multiple utility scales within the study to assess quality of life. EQ5-D and SF-6D were the most commonly used utility scales. Using a random effects model we observed a change in utility per unit increase in BMI of -0.0054 [-0.0077; -0.0031]. However there was significant heterogeneity between studies. The number of covariates ranged greatly between the studies and appeared predictive of the magnitude of effect of BMI on utility. CONCLUSIONS We illustrate methods for meta-analysing multiple parameter estimates and discuss the selection of appropriate parameter estimates for the inclusion in cost-effectiveness models. In particular we illustrate the relationship between the selection of appropriate parameter estimates in terms of which covariables were included in the originating studies and the cost-effectiveness model structure in terms of which independent causal effects are modelled.