Abstract: The droplet initial size model constitutes a fundamental element of subchannel analysis for the description of droplet diameter, transport behavior, and interfacial heat transfer, and it has a substantial impact on the prediction of downstream thermal-hydraulic parameters under reflood conditions. Accurate representation of entrained droplet behavior is particularly important for post-critical-heat-flux heat transfer analysis because it directly affects the prediction of liquid cooling performance and safety-related parameters such as cladding temperature. In this work, the effects of spacer grid modeling and droplet initial size correlations on reflood prediction were investigated using the two-fluid three-field subchannel code. The influence of spacer grid representation on the accuracy of reflood parameter prediction was examined by comparing different modeling approaches. The results indicate that the equivalent resistance treatment, in which the spacer grid is represented solely by its pressure-loss effect, is inadequate for reflood simulations. Although this simplified approach accounts for the hydraulic resistance introduced by the grid, it cannot reproduce the enhancement of downstream heat transfer or the droplet breakup process induced by the grid structure, leading to noticeable deficiencies in the prediction of downstream flow and heat transfer characteristics. These findings imply that a realistic simulation of reflood phenomena requires a more mechanistic treatment of spacer grid effects. On this basis, six post-critical-heat-flux initial entrained droplet size models, namely the RELAP5, COBRA-TRAC, Paik, TRACE, Analytis, and Yoo models, were implemented into subchannel code, and their influences on major thermal-hydraulic parameters during the reflood process were systematically assessed. The comparative results reveal that the selection of the droplet initial size model leads to significant differences in the predicted droplet Sauter mean diameter, droplet velocity, and cladding temperature, and these discrepancies further propagate to the prediction of droplet transport, interfacial heat transfer, and wall cooling behavior in the post-CHF regime. Among the models considered, the Analytis model exhibits the best overall agreement with the experimental data, providing comparatively accurate predictions of droplet Sauter mean diameter, the variation trend of droplet velocity, and peak cladding temperature. By contrast, the RELAP5 and Yoo models tend to overestimate the initial droplet diameter, which results in lower predicted droplet velocities and reduced cooling effectiveness of the dispersed liquid phase, thereby causing a marked overprediction of cladding temperature. The present study demonstrates that the droplet initial size model plays a decisive role in determining the predictive capability of subchannel codes under reflood conditions. A physically appropriate and well-validated droplet initial size model can significantly improve the simulation of droplet dynamics and cladding cooling behavior, whereas an unsuitable model may introduce considerable errors in the prediction of important safety parameters. It is necessary to conduct a detailed assessment of the applicability of droplet initial size models during the development and application of subchannel codes.