Structural bonding is nowadays widespread in the industry. However, characterisation methods and 3D modelling of the adhesives need to be improved. The characterisation requires an experimental procedure to obtain a large experimental database under various loading cases, which represents a significant amount of data. The 3D modelling requires advanced models with several parameters to identify and generally uses inverse identification procedures, which can be time expensive. For a good accuracy, the constitutive models need to take into account the dependency on the hydrostatic stress and be written under the non-associated formalism. In this study, the experimental database is obtained via a modified Arcan test that can cover a wide range of loadings between tension, shear, mixed tension-shear, and mixed compression-shear. A second experimental campaign is realized with a tension/compression-torsion (TCT) test that can cover a greater range of loadings: from tension to compression and mixed tension/compression-shear, with an infinite possibility of mixed loadings. The modified Arcan database is used to identify a 3D elastic-plastic Mahnken-Schlimmer type model, according to an inverse identification procedure developed in a previous study. This model identification is validated on the experimental database coming from the TCT test: a numerical/experimental comparison is realized. This allows the validation of the model and emphasizes the benefits of the TCT test. Indeed, it proves that this test is well suited to characterize adhesive joints and presents several capacities that will be really useful for further studies, like an infinite range of non-proportional loadings available.