This study aims to better understand the impact of different building representations and mesh resolutions on urban flood simulations using the TELEMAC-2D model in idealized urban districts. A series of numerical models based on previous laboratory experiments was established to simulate urban flooding around buildings, wherein different building layouts (aligned and staggered) were modeled for different building representations: building–hole (BH), building–block (BB), and building–resistance (BR) methods. A sensitivity analysis of the Manning coefficient for building grids indicated that the unit-width discharge and water depth in building grids reduce as the Manning coefficient is less than 104 m-1/3⋅s. The simulated depths via the BH, BB, and BR methods were compared with the measured data in terms of three accuracy indicators: root mean square error, Pearson product–moment correlation coefficient, and Nash–Sutcliffe efficiency. Observing apparent discrepancies based on the hydrographs was difficult; however, some slight distinctions were observed based on the aforementioned three indicators. The sensitivity of 1, 2, and 5 cm mesh resolutions was also analyzed: results obtained using 1 cm resolution were better than those obtained using other resolutions. The complex flow regime around buildings was also investigated based on mesh resolution, velocity, and Froude number according to our results. This study provides key data regarding urban flood model benchmarks, focusing on the effect of different building representations and mesh resolutions.