<p>On April 3, 2024, a magnitude <i>M</i>_w 7.4 earthquake struck the city of Hualien, Taiwan, China, causing casualties and immense damage. This earthquake triggered a tsunami, which was recorded by the nearby tide gauges and Deep-Ocean Assessment and Reporting of Tsunamis (DART) buoys. These recordings are valuable for evaluating the tsunami source and quantifying the tsunami characteristics. In this study, we conduct tsunami simulations based on three earthquake source models, and analyze their reliability by comparing the computed results with observed waveforms. The evaluated source models are the United States Geological Survey (USGS) finite-fault model, and two uniform slip models using different scaling relations. The tsunami waves generated by each source are simulated with a non-hydrostatic tsunami model, which accounts for the effects of wave dispersion. The computed tsunami arrival times and wave heights are compared to the observed data and show high consistency, indicating that the magnitude and location of the earthquake source are well estimated. Also, the three source models with different rupture area and average slip lead to almost the same tsunami waves at each station, which suggests that earthquake rupture details have limited impact on far-filed tsunami records. It is also found that wave dispersion effects in this event are negligible at most stations. The findings are useful for tsunami warning. For fast warning purposes, it is practically useful to adopt simplified uniform slip models, which are able to predict the tsunami arrival time and wave height relatively well without knowing the earthquake source details.</p>