Pulse‐like ground motions cause severe damage in structures at certain periods. Hence, pulse effects need to be considered during probabilistic seismic hazard analysis and seismic design in the near‐fault region. Traditional ground‐motion models used to quantify the hazard posed by pulse‐like ground motions may underestimate them, but they are relatively suitable for describing the residual ground motions after extracting pulses. Nevertheless, the applicability of Next Generation Attenuation‐West2 Project (NGA‐West2) models to pulse and residual ground motions has not been evaluated. Moreover, the applicability of recently developed directivity models, including the Shahi and Baker (2011; hereafter, SB2011), Chang et al. (2018; hereafter, Chang2018), and Rupakhety et al. (2011; hereafter, Rupakhety2011) models, has not been investigated for this event. Here, based on the abundance of pulse‐like ground motions recorded during the 6.4 Hualien earthquake, the applicability of NGA‐West2 models and directivity models was quantitatively evaluated. In summary, (1) The applicability of NGA‐West2 models to the observed original and residual ground motions varies significantly at different periods. The suggests that NGA‐West2 models overestimate the original and residual ground motions for short periods (), but are suitable for describing the residual ground motions yet underestimate the original ground motions for long periods (). (2) Pulse periods and amplification bands due to pulses are unusually larger than previous events. Similar to the Chang2018 model, the plateau of this event starts and ends at the periods of 0.70 and 1.1 times the pulse period. However, the Chang2018 and SB2011 models underestimate the constant ordinate of this plateau. Spectral ordinates of the spectral shape curve due to pulses for the short period () are smaller than the predictions from the Rupakhety2011 model. The trend was reversed for long periods (). Compared with the Rupakhety2011 model, the peak location of the spectral shape curve is shifted to the long period. These results will be helpful for updating these models in the near future.