Background: Although the goal of meta-analyses is to summarize existing knowledge and improve patient care, little is known about how authors and readers actually interpret them. We recently demonstrated that reviewers interpret the same meta-analysis very differently. Objectives: We are conducting a mixed-methods study to characterize convergences and divergences between how meta-analysts believe they interpret meta-analyses and how they actually do interpret them. Our first step reported below was to develop and pilot test a quantitative questionnaire that will be used to guide individual interviews. Methods: Each participant provided general information about themselves, and outlined how they approach meta-analyses including likelihood functions and prior beliefs about potential effectiveness and variability of new treatments for medical conditions. We then showed each participant a simulated meta-analysis based on three ‘perfect’ RCTs and asked them to provide an overall effect estimate. We then repeated this after adding two RCTs (five in total), five more RCTs (10 total) and eight more RCTs (18 total). We used a common Bayesian approach and compared the calculated posteriors based on reported priors and likelihood functions to their reported posteriors. Participants provided feedback and the questionnaire was adapted after each participant until no further substantive recommendations were made. Results: Moderate changes were made after the first three participants, but the fourth and fifth participants provided only minimal suggestions and we therefore stopped pilot testing at this point. The preliminary analyses suggest that most participants are not interpreting data consistent within the usual Bayesian framework of priors, likelihood functions and posteriors (Figure 1). It appears that participants either have difficulty providing appropriate priors and posteriors, or that the standard Bayesian models fail to account for some aspect of the cognitive process. Conclusions: Participants may interpret meta-analyses differently from how the Bayesian models indicate they should be interpreting them. Funding: Canadian Institutes of Health Research.