To analyze the delamination behavior of a thermoset composite, the parameters of a mixed mode cohesive model are identified by combining experimental displacement fields measured via global digital image correlation and load data in a single cost function written within a Bayesian framework using Finite Element Model Updating. A sensitivity analysis shows that the load data are of prime importance to enable the sought parameters to be tuned. Two delamination tests are analyzed in a single procedure that is first decoupled to find good initial guesses, and then fully coupled to get a set that is compatible with both experiments. The sensitivity of the calibrated parameters to their initial guess is analyzed. A low model error is found thereby validating the proposed approach (enabling to choose the best strategy for identification steps).