Background: Several meta-analysis models for combining diagnostic test data have been described in the literature (Deeks 2001, Harbord 2007). These models vary in the assumptions they make regarding the (i) variability in test thresholds between studies and (ii) incorporation of variability beyond that expected by chance (between-study heterogeneity). In any particular situation, it is unclear which model is the most appropriate for the data. Traditionally, goodness of fit criteria are used to choose between different statistical models. Where complex nonnested models with random effects are considered (as in this situation), the Deviance Information Criteria (DIC) (Spiegelhalter 2002) provides a criteria for choosing between models. Objectives: To explore the use of DIC to choose between different meta-analysis models applied to 198 studies evaluating DDimer for deep-vein thrombosis (DVT) (Goodacre 2005). Methods: To meta-analyse the DVT diagnostic test data, the following random/fixed effect models are fitted: 1. independent estimates of sensitivities/specificities; 2. symmetric summary-ROC curves estimation; 3. asymmetric summary-ROC curve estimation; 4.bivariate estimate of sensitivities/specificities. In addition, these models are extended to include covariates. The fit of the different models is assessed using the DIC. Results: As can be observed in Table 1, the Bivariate model fits best for this example (DIC:2133), and the fit is improved by including a covariate for mean age of patients (DIC:2120). The mean age of the study population in the bivariate model affects either sensitivity [+0.039 increase per year, 95%CI (-0.008 to 0.087)] or specificity [ -0.040 increase per year, 95%CI (-0.074 to -0.006)]. Conclusions: With numerous alternative approaches available for meta-analysis of diagnostic test accuracy data, each making different assumptions, a way of choosing between models is required. The use of DIC seems to be well suited.