In the first part of this study, a set of 87 ground‐motion records, with closest distance to the rupture plane (Rrup) less than 200 km and averaged shear‐wave velocity over the top 30 m of the subsurface (VS30) between 175 and 1400 m/s, recorded during the 2012 Ahar–Varzaghan dual earthquakes (Mw1 6.4, Mw2 6.3) were taken into account to examine the predictive capabilities of the Next Generation Attenuation (NGA) ground‐motion prediction equations (GMPEs) via a set of comparative analyses and tests. The first‐applied method to assess the performance of the NGAGMPEs is based on the intraevent residual analysis. The primary database (i.e., 87 records) was also used to develop an event‐specific GMPE in the case of the Ahar–Varzaghan dual earthquakes by means of regression analyses. The derived event‐specific GMPE has been compared with the NGAGMPEs for two different site conditions, that is, VS30>375 m/s (rock site) and VS30>375 m/s (soil site). The residual analysis results indicate that the NGAGMPEs perform better in predicting data recorded at rock sites compared to soil sites. For soil sites and at large periods (T=2.0 s), the observed spectral accelerations are overpredicted by the NGAGMPEs. Furthermore, as the second part of this study, to select the most adequate GMPEs, 14 strong‐motion records from the 1997 Ardebil earthquake (Mw 6.1) were added to the primary database. The implementation of the likelihood (LH) and log‐likelihood methods, as modern LH‐based ranking assessment techniques, as well as the Nash–Sutcliffe index reveal that the NGAGMPEs show good compatibility at short‐medium periods (T<1.0 s) on the basis of the data recorded during the 2012 Ahar–Varzaghan dual and 1997 Ardebil earthquakes (i.e., 101 records). However, in the long‐period range, the dispersion in the data does not allow the authors to draw a comprehensible conclusion.