•3D fracture loci of uncoupled ductile fracture models were calibrated for 2024-T351 aluminum alloy and DP600 steel.•The type and number of calibration tests has a marginal influence in the fracture locus and predictive ability of the models.•Quite different shapes of the fracture locus are observed for between both materials.•The models present different levels of prediction accuracy in the high range of stress triaxiality, for both materials.•It is difficult to select the most appropriate model based only on the objective function value. The prediction of ductile failure is crucial for the progress of metal forming industries. Therefore, uncoupled fracture models have been proposed and continuously extended, trying to improve their forecasting abilities. However, without an assessment analysis of their predictive ability, it is difficult to establish the most suitable model, for describing the onset of failure under a wide range of loading conditions. Actually, a fair comparison of models’ predictive ability requires the use of an extensive experimental database and a similar calibration procedure. In this context, seven uncoupled fracture models are chosen and revisited to study the effect of the type and number of tests used for calibration on their predictive ability. Experimental datasets available in literature were considered for 2024-T351 aluminium alloy and for DP600 steel. For each material, the experimental dataset was split in two subsets. The first one includes tests that were carried out under plane stress conditions that were used in the calibration procedure. For the aluminium alloy, different groups of 6 tests, covering distinct stress states, were elicited from the experimental subset and applied in the calibration of the seven models. The remaining tests were used to assess the predictive ability and their sensibility regarding the group used in models’ calibration. Different trends of the fracture loci are observed, depending on the selected models and the tests used in their calibration. However, the analysis of the results indicates that a set of at least 6 experimental tests is appropriate to calibrate fracture models, providing that they enable covering a wide range for the stress triaxiality and Lode parameter. Finally, the conclusions provide some recommendations for the improvement of the calibration procedure.