Here, the generalized inversion of S‐wave amplitude spectra is employed for deriving the site response and S‐wave attenuation (quality factor) in the northwest region of Iran. A total of 279 strong‐motion accelerograms recorded at distances below 200 km from 41 earthquakes with moment magnitudes ranging from Mw 3.8 to 6.5 are used to determine the region‐speciﬁc attenuation model. The site responses have been determined for all 54 stations individually. The bulk of strong‐motion data (i.e., 184 records) comes from the 2012 Ahar–Varzaghan dual earthquakes in northwest Iran and their aftershocks. Also, the modified finite‐fault method, which is able to model nonuniform stress distribution on the fault plane, is employed to investigate the coseismic stress parameter for the first and second Ahar–Varzaghan dual earthquakes by minimizing the difference between the synthesized and observed pseudospectral accelerations for 5% damping. Results denote a concentration of stress drops on the northwestern part of the first event’s fault and on the central and eastern part of the second fault. Stochastic simulations, using the calibrated stress distribution have been performed on a regular grid covering the study area for both rock‐ and soil‐site conditions (in total, 651 points). The results show that the maximum of peak ground accelerations (PGAs) during the Ahar–Varzaghan dual earthquakes for the rock‐site condition reached 557 cm/s/s, and the largest peak ground velocities (PGVs) were estimated around 76 cm/s. In general, comparisons between simulated data and empirical ground‐motion prediction equations (GMPEs) show that the stochastic predictions (PGAs and PGVs) are higher than those predicted by the empirical GMPEs for distances less than 30 km and are smaller for distances more than 30 km. The simulated data are also consistent with the modiﬁed Mercalli intensity observations and show reasonable agreement. This can be considered as another validation of the method.