Presented herein is a new set of empirical models for prediction of horizontal peak ground velocity (PGV), peak ground acceleration (PGA), and 5% damped elastic response spectra in the 0.01–10 s vibration period range in Iran. The models are based on a carefully curated databank of Iranian strong ground motions composed of 1350 pairs of orthogonal horizontal acceleration time histories from 370 regional earthquakes with and source‐to‐site distances up to 200 km. The adaptive wavelet denoising approach used for waveform processing in this study allows extending the range of oscillator periods for which predictions can be made. We investigated the possible regional dependency of strong ground motions from three different regions (Alborz–Azerbaijan–Kopeh Dagh, Zagros, and central Iran–Makran) of the Iranian plateau through analysis of variance. Our results show that data from these regions can be combined into a single database because only a few magnitude–distance intervals exhibit significant regional differences. Local site effects are taken into account using a site classification scheme based on . The predictive models also include style‐of‐faulting (SoF) terms. Because of the paucity of data for normal‐faulting events, the applicability of the proposed attenuation models is limited to unspecified, strike‐slip, and reverse SoFs. We propose four models that employ point‐source (hypocentral and epicentral) and extended‐source (rupture and Joyner–Boore) distance metrics to enhance the flexibility of the prediction tool within the context of seismic hazard assessment studies. The results of between‐event and within‐event residuals show that the proposed new models are unbiased. We present a detailed comparison of our results with a selection of local, regional, and global predictive models developed by other authors.