Abstract: During the refill phase of a large break loss-of-coolant accident, a large amount of steam generated in the core region flows into the downcomer through the lower head. The counter-current flow limitation (CCFL) phenomena may occur in the downcomer region of the reactor pressure vessel when subcooled water is injected from the safety injection system. This CCFL effect can reduce the effective injection volume of the RIS due to water bypass. To characterize the CCFL behavior, an RPV integral test facility was designed and constructed based on scaling criteria and design features of prototype reactor. The experimental investigations were conducted. Subsequently, the condensation characteristics were quantified and different CCFL models were evaluated. Experimental results indicate that intense energy exchange occurs between safety injection water and steam. Under low steam flow conditions, the subcooled injection water achieves full saturation through complete heat transfer with the steam phase. However, steam velocity affects condensation capability. As steam velocity increases, condensation efficiency decreases. Still, the minimum efficiency remains significant. It stays at 0.86 across all experiments. Based on the experimental results under different operating conditions, fitting and mechanistic analysis were conducted to obtain the specific intercepts corresponding to the two different CCFL models. Numerical simulations conducted using system analysis code with these two experimentally derived CCFL models. Then, the numerical simulations were performed. The results reveal a significant finding. The computational model consistently underestimates steam condensation rates. This underestimation specifically affects condensation characteristics of safety injection water. Consequently, the simulation overpredicts water loss through the break. The differences between numerical and experimental results are systematically examined. The applicability of system analysis code in this context is critically assessed. Finally, the robust model selection methodology is proposed. This approach aims to improve prediction accuracy.