The flow through meander bends is inherently three dimensional and may be characterized by primary flow in streamwise direction and secondary flow in transverse direction. Although, three dimensional simulations of meander bends are feasible for laboratory scale experiments, temporal and spatial scales of naturally occurring meandering rivers are much larger than those found in laboratory experiments. Therefore, computationally less expensive (reduced) flow models are necessary. Reduced hydrodynamic models are depth-integrated models and therefore require a closure model to resolve the effect of secondary flow. At present, most reduced flow models are linear as they neglect the feedback between the primary and secondary flow, which limits their validity to mild curvature. A non-linear reduced flow model, including this feedback, is compared to experimental data from the laboratory and the field which are both sharply and moderately curved. The model predictions compare well to the global flow structure in the high curvature Kinoshita flume and the moderate curvature Tollense River bend (with extra complicating factors of vegetation and horizontal recirculation zones). A linear model is also used to model the selected cases, showing good agreement for moderate curvature but not for sharp curvature. An analysis of the driving mechanisms reveals the reason for the difference in the model predictions.