Background <p>Determining accurate constitutive laws from experimental data remains a key challenge in mechanics, particularly when the material behavior is nonlinear and the dataset is limited or noisy. Traditional approaches rely on identifying parameters of preselected material models, which separates the model selection and the calibration tasks leading to a potentially long and tedious trial-and-error procedure.</p> Objective <p>This work aims to assess the performance of EUCLID (Efficient Unsupervised Constitutive Law Identification and Discovery), a recently proposed framework for automated discovery of constitutive laws, when applied to experimental data.</p> Methods <p>Mechanical tests are conducted on natural rubber specimens of varying geometrical complexity. Both global (force–elongation) and local (full-field displacement) data are collected. Constitutive laws are obtained via two routes: (i) conventional parameter identification in a priori selected models, and (ii) EUCLID, which integrates model selection and parameter identification in a unified model discovery pipeline.</p> Results <p>The two approaches are compared in terms of predictive accuracy, generalization to unseen geometries, and robustness to experimental noise. The coverage of the material state space achieved by each dataset is quantified, and the relative performance of different datasets and models is analyzed.</p> Conclusions <p>EUCLID enables automated and data-driven discovery of constitutive laws, offering improved flexibility compared to conventional identification methods, and showing strong potential for reliable material modeling based on experimental data.</p>