Background: Because predictive values tend to vary more with changes in prevalence than other measures of diagnostic accuracy, direct meta-analysis of positive predictive values is discouraged. However, it is unclear whether meta-analyzing predictive values directly will indeed produce different results than first obtaining summary estimates of sensitivity and specificity and calculating predictive values by Bayes theorem. Objective: To compare the conventional bivariate logitnormal model for meta-analysis of test accuracy studies, which results in summary estimates of sensitivity and specificity, with an alternative bivariate logitnormal model of the positive and negative predictive value. Methods: From a set of meta-analyses that included a consecutive series of eligible patients, we estimated a summary sensitivity and specificity for each review, using the bivariate method for meta-analysis. Then we used the mean prevalence to calculate predictive values. The bivariate model was also used to directly obtain summary estimates for predictive values. The final estimates for predictive values were compared, as well as heterogeneity measures and the −2 log likelihood to compare how the models fitted the data. Results: Sixteen reviews fulfilled our criteria and were analyzed with both models. Of these 16 reviews, 10 showed a lower −2LL for the predictive values model, and 6 showed a lower −2LL for the conventional model. The estimated predictive values did not differ significantly. Discussion: Our results do not show a significant preference for either the conventional model or the direct model for predictive values. Although predictive values may be preferred by clinicians, the question remains whether, for example, the effect of covariates on predictive values can be interpreted the same way as their effect on sensitivity and specificity. Recently, a trivariate model of sensitivity, specificity and prevalence was published. We will compare the above models with this model as well and present those results during the Colloquium.