Recent studies have shown that repeatable travel‐path terms make a high contribution to the overall variability in earthquake ground motions. Having maps of such terms available for a given recording site would, theoretically, allow removal of this component from the aleatory variability of ground‐motion models. The assessment of such travel‐path terms for a given site, however, relies on having recorded a rich set of earthquakes at that site. Given the relative youth of strong‐motion networks, the assessment of such terms from observations is currently difficult for most parts of the world. Ground‐motion simulations provide an alternative method to assess such terms. In this article, many dozens of earthquakes, distributed in a grid, are simulated for the Marmara Sea region (Turkey), which borders the megacity of Istanbul and is an area of high seismic hazard. Ground motions are simulated within a detailed 3D velocity structure model using a finite‐difference method at 70 recording sites in the area (200×120 km2). Horizontal peak ground velocities from these simulations are regressed to derive a ground‐motion model. Next, residuals from this ground‐motion prediction equation are computed to assess repeatable source, site, and path terms and various components of ground‐motion variability. These components are similar to those derived from real strong‐motion data, thereby lending support to those estimates as well as showing the worth of simulations for this type of exercise.