We simulated broadband hard-rock ground-motion time histories for the Saint Louis, Missouri, metropolitan area due to three large earthquakes (6.5 ≦ Mw ≦ 7.5), each occurring on the largest fault segment of the New Madrid fault zone at a distance of about 200 km. A deterministic procedure was implemented to simulate ground motion for frequencies less than 1 Hz for which the propagation effects were calculated using full-wave theory. For signals above 1 Hz, a semi-empirical simulation approach was used. The fault dimensions were defined in conformity with a constant stress-drop model developed for eastern North America; the fault width was constrained to a maximum depth of 16 km following the occurrence of the natural seismicity. For each event, we simulated broadband time histories for ten randomly generated slip models at eight sites distributed in various locations in the city of Saint Louis. This study shows that the level of peak ground motions at hard-rock sites in Saint Louis is small; the largest average peak horizontal ground-motion acceleration simulated is only 26.45 cm/sec2 (1.96 cm/sec for the velocity) even for the largest event. To validate the simulation results, we examined regional accelerograms and their peak values and response spectra recorded during the 25 November 1988 Saguenay earthquake (Mw = 5.8) at distances further than 230 km from its epicenter to be consistent with the distance of the city of Saint Louis from the New Madrid earthquake zone. The Saguenay earthquake observations seem consistent with simulation values obtained for the Mw = 6.5 New Madrid earthquake. A preliminary equivalent linear analysis by applying these simulated motions at the base of a soil profile suggests that the level of ground motions is too weak to produce damaging effects in Saint Louis even for the largest event on the longest segment.