Developments in diagnostic tools and an increase in computational power allow for extremely detailed digital twins in the analysis of historical masonry construction. However, experimental validation of the necessary level of detail for reliable safety assessment has been scarce, partly because experiments are focused on the global failure patterns and rely on small-scale models. This paper presents a comparative analysis of the failure of masonry arches on spreading supports. It contrasts the results of an experimental campaign using large-scale models and discrete element analysis. For their relevance, three arch geometries, semi-circular, equilateral, and obtuse pointed, are considered. The study focuses on the development and change in the damage pattern while the arches are loaded to failure. The combined (experimental and numerical) results confirm that while small geometrical imperfections can influence the local failure pattern, they have little effect on the overall load-bearing capacity of the structure. A discrepancy between the local failure in the DEM model and the physical model is also noted, which contributes to understanding the limitations of discrete element analysis. Notably, the maximum displacements recorded during the experiment are markedly higher than previous analytical studies suggested, which deserves further study.