The internal designs of two tundish nozzles A and B to deliver liquid steel in a slab mold are characterized by a full‐scale water model and the scale‐adaptive simulation (SAS) turbulence model. The internal flows in the nozzles determine the flow patterns in the slab mold and their clogging tendency. The SAS model predicts well the sub meniscus velocities measured through ultrasound signals and the unsteady flows of a given nozzle. The turbulence structures in the sub‐meniscus region yield slow descending rates of velocity–time autocorrelations leading to flow turnovers of low frequency using both nozzles. However, nozzle A (a bore with one expansion and one contraction) yields larger power density spectra than nozzle B (designed with internal flow deflectors) in the sub meniscus region, indicating that this latter nozzle dissipates more the turbulent energy, effect due to its internal flow deflectors. Four clogging criteria, wall‐shear, Q, turbulence eddy frequency, ω = ε/k, and the instantaneous velocity in the boundary layer, together with the flow structure, prove to predict the propensity of a given nozzle design to clog by alumina inclusions in a particular region. This analysis reveals that nozzle B has a lower tendency to suffer clogging during industrial operation.