Using 72 free‐field accelerograms with the closest site‐rupture distances () less than 300 km, this study calculates the recorded significant duration of the 2013 6.7 Lushan earthquake and compares the calculated results with empirical model predictions. The significant duration parameters are for 5%–75% () and 5%–95% () of the normalized Arias intensity. For , the models of Du and Wang (2017; hereafter, DW17) are relatively less biased in predicting the Lushan significant duration. The goodness‐of‐fit results indicate that the models of DW17, Afshari and Stewart (2016; hereafter, AS16), and Kempton and Stewart (2006; hereafter, KS06) are the best models to predict of the Lushan event. Meanwhile, most of the selected empirical models predict the Lushan event better for than for . Moreover, the enrichment of high‐frequency components of ground motion causes the recorded of the Lushan event to be larger than median predictions in high‐frequency bands. Among the five selected earthquakes (1994 6.7 Northridge, 2003 6.5 San Simeon, 2004 6.6 Niigata, 2008 7.9 Wenchuan, and 2013 6.7 Lushan earthquakes), the Lushan earthquake is predicted relatively better by all empirical models. The intraevent and interevent residuals of AS16 model from China earthquakes are within the scatter of those from the events in the Next Generation Attenuation‐West2 (NGA‐West2) database, indicating that the China earthquake significant duration is consistent with the model developed through the NGA‐West2 database. Furthermore, the regional dependences of significant duration are not observed for the Wenchuan or the Lushan earthquake.