The original aim of the present study was to test the efficiency of some selected ground‐motion prediction equations (GMPEs) against small‐to‐moderate data recorded in the Iranian plateau. For this purpose, we applied three statistical tests including the Nash–Sutcliffe model efficiency coefficient (Nash and Sutcliffe, 1970) and likelihood and log‐likelihood methods (Scherbaum et al., 2004, 2009) to assess performance of eight GMPEs against a comprehensive databank from small‐to‐moderate magnitude data in Iran. Two of the candidate models were selected from Japanese models, and the rest were chosen from regions with shallow crustal earthquakes including local models corresponding to the Iranian earthquakes, regional models established for the Middle East and Euro‐Mediterranean regions, and global models (developed for shallow crustal environments). We analyzed a high‐quality dataset composed of 937 ground‐motion records from 296 Iranian earthquakes with moment magnitude ranging between 3.9 and 5.0 and epicentral distances of up to 100 km. In conclusion, Zafarani et al. (unpublished manuscript, 2015; see Data and Resources) as a ground‐motion model valid for a magnitude range as small as M 4.0 is the only model that shows good consistency with the recorded data over all frequencies. This emphasizes the importance of region‐dependent anelastic attenuation and the necessity to develop more elaborate local GMPEs for small‐to‐moderate earthquakes, which are also very important from the viewpoint of epistemic uncertainty.