Various efforts have been made to improve the models and correlations of RELAP5/MOD3.3, which are activated only during the reflood phase of a large break loss-of-coolant-accident, the so-called "reflood model". The film boiling heat transfer regime was divided into three sub-regimes and appropriate wall-to-fluid heat transfer coefficient correlations were applied to each sub-regime. The models for estimating the size and the velocity of droplets were also modified. In addition, we introduced a special model to take into account the wall-to-fluid or interfacial heat transfer enhancement due to the spacer grids. The modified reflood model was assessed against eight FLECHT-SEASET tests and six RBHT tests having various initial and boundary conditions. The modified reflood model better predicted the PCTs and quench times of the eight FLECHT-SEASET tests: the RMS deviation of PCT predictions was reduced from 51.7K to 29.8 K and it was reduced from 81.9 s to 33.5 s in the case of quench time predictions. As for the six RBHT tests, the quench times were predicted much better by the modified reflood model while not much improvement in the PCT predictions was obtained: the RMS deviation in quench time predictions was reduced from 322.5 s to 132.1 s while the RMS deviation of PCT predictions were 62.5 K and 53.1 K before and after the modifications, respectively. Additional calculations were conducted excluding the spacer grid heat transfer enhancement model to reveal that the model is essential to predict correctly the cladding temperature behaviors in the upper part of the core. (C) 2012 Elsevier BM. All rights reserved.