This paper examines the influence of material modeling choices on the behavior of soil–steel arch bridges, which are defined by their complex soil–structure interaction. The performance of these bridges is highly dependent on the stress history, particularly during the backfilling stage and under subsequent live loads. Accurate modeling, therefore, requires the use of an appropriate constitutive model for the backfill to reliably capture its response under varying loading conditions. In this study, a detailed three-dimensional, finite element model of a 14-m-span soil–steel arch bridge was developed. The analysis incorporated the effects of soil compaction during the placement of successive backfill layers around the flexible culvert. The effectiveness and suitability of three different soil constitutive models were systematically evaluated. The results show that the Mohr-Coulomb model overestimates the footing's average settlement, lateral displacement, and differential settlement by 96.8%, 113.1%, and 59.3%, respectively, at the end of the backfilling process when compared to the Hardening Soil model with Small-Strain Stiffness. Additionally, the Mohr-Coulomb model underestimates the final thrust force at the crown by 4.9% and at the base by 7.9%.