The empirical Green’s functions (EGF) technique is used to investigate two methods for predicting ground motion in a sedimentary basin for a future earthquake, including variability assessment. This study focuses on the Grenoble basin (French Alps). The basic principle of both methods is to generate a variety of source parameter sets based on a grid testing approach. Next, these sets are used to compute a population of ground motions by means of a kinematic EGF method and to estimate ground-motion variability. The first method tried, called the direct-parameter-input approach, selects input parameter combinations from assumed source parameter probability density functions. It is demonstrated that this approach leads to overestimated variability. Moreover, these simulation results are not calibrated. A new (screened-parameter-input) procedure is therefore proposed: (1) reference rock site response spectra are simulated for fractiles of several orders using empirical ground-motion prediction equations; (2) a large population of rock site response spectra is generated by means of the EGF method with varying rupture parameter combinations; (3) the spectra that do not fit the empirical motion for the chosen fractiles are screened out and sets of permissible source parameter combinations are thus obtained; (4) sediment site response spectra are computed with this EGF procedure and with these permissible parameter combinations; and (5) for each frequency median spectral acceleration and standard-deviation values are derived.