The reflood model of RELAP5/MOD3.3 was assessed using eight carefully selected FLECHT-SEASET tests. Comparisons of calculated and measured peak cladding temperatures and quench times indicate that the code predicts peak cladding temperatures relatively well. However, rod quenches were predicted to occur too early. To improve the predictability for quench times, we carefully reviewed and modified the wall-to-fluid heat transfer models. The film boiling heat transfer regime was divided into three different sub-regimes, and appropriate correlations were selected for the wall-to-fluid heat transfer coefficients in each sub-regime. We modified also the wall-to-liquid radiation heat transfer model including the method for determining the droplet diameter and we corrected two minor coding errors existing in the original version. After the modifications, the same set of eight FLECHT-SEASET tests was simulated again to compare the newly predicted cladding temperatures, quench times and heat transfer coefficients with those predicted using the original version. The modifications employed in this study improve the code's predictability not only for quench times but also for peak clad temperatures. The modifications reduced the Root-Mean-Square (RMS) error in the prediction of peak clad temperatures and quench times from 48.3 to 36.1 K and 85.9 to 33.2 s, respectively. (C) 2009 Elsevier B.V. All rights reserved.