The work proposed in this paper aims at improving the physical understanding and modeling of junction flows by conducting an analysis of several quadratic constitutive relations (QCRs) applied to corner vortices. Reynolds-averaged Navier–Stokes computations are performed on two academic configurations for verification purposes and on a technical application of wing–body juncture featuring corner flows to achieve a better comprehension of the QCR variants’ ability to reproduce turbulent stress combinations relevant to secondary flows generation. This addresses the more theoretical interest of analyzing the onset of corner separation (at an angle of attack [Formula: see text]) and the role of corner flow vortices. Linear eddy viscosity model and the extended QCR failed to detect these vortical structures; the latter encountered numerical stability difficulties. On the other hand, QCR2000, QCR2020, and QCR(r) showed behavior consistent with a simulation using the Reynolds Stress Model, experiments, and literature results. This better prediction was found to be related to the ability of these models to accurately predict the stress-induced vortex that acts in delaying and reducing the size of the corner separation. Notably, the more recent QCR(r) and QCR2020 have enhanced the estimation of the normal stress combination [Formula: see text] compared to the more widespread QCR2000, demonstrating their validity as alternative models.