We simulated the strong-motion time histories recorded during the main shock (Mw 6.0) of the Umbria-Marche seismic sequence (Central Italy) of September–October 1997. Ground-motion waveforms were computed using the stochastic modeling technique proposed by Beresnev and Atkinson (1997, 1998) for finite faults. In this approach, the high-frequency amplitudes are simulated as a summation of stochastic point sources. We used the FINSIM code (Beresnev and Atkinson, 1998), which incorporates regional attenuation and frequency-dependent site-amplification factors. We divided the fault plane into 60 elements whose length and width are 1.2 km and 1.5 km, respectively. The resulting subfault corner frequency and rise time are 0.91 Hz and 0.5 sec. We found that the site-amplification functions play an important role in the simulation process, improving the fit to the observed time histories and spectra. The strong-motion waveforms recorded at the Nocera (NOC) station, located at the northern end of the causative fault, show an important directivity effect. Thus, to fit the observed ground motions, we used an inhomogeneous slip distribution, weighting the slip on the fault heavier toward the north. We tested two models: one that simulates a fault rupture with two main slip patches and a second model that ruptures northward with a less heterogeneous slip distribution where slip is mostly concentrated near the rupture nucleation at the southern edge of the fault plane. The simulated low-frequency amplitudes at NOC, however, require an additional frequency-dependent directivity correction (e.g., Bernard et al., 1996). In conclusion, we found that stochastic finite-fault simulations calculated using adequate site amplification functions and crustal attenuation reproduce reasonably well the ground motions from the Mw = 6.0 Umbria-Marche earthquake.