We examined how variations in bathymetric resolution affect simulated coastal tsunamis. We used 1-m resolution LiDAR datasets to generate resampled bathymetries with 5, 10, 20, 30, 40, 50, 100, 200, and 300 m resolutions, and added GEBCO data in the comparison. Tsunami waves offshore were generated by setting up an instantaneous rupture sourced from a hypothetical fault model. We used the COMCOT software to model tsunami propagation towards the coas. Using the 5 m bathymetry as the reference model, we observed that datasets with 10 – 50 m resolutions can reproduce tsunamis reasonably well. The maximum heights are overestimated by ≤5% or underestimated by ≤10%, and the first wave arrival time is ∼10%earlier than expected. Coarser bathymetries show an increasing trend of height underestimation, with the GEBCO model underestimating it by as much as 70%. Coarser bathymetries have more variable first wave arrival time, either ≤20% later or ≤10% earlier. Overall, a reasonably accurate result can be achieved using a bathymetric resolution in the 10 m – 50 m range, and is achievable with reasonable computational efficiency. This study highlights the importance of shallow bathymetry in the numerical modeling of tsunami propagation.