Stochastic simulations have recently become quite popular for estimating synthetic ground motion time histories. For seismically active regions that are not well-monitored or studied extensively, input parameters of the simulations should be carefully selected as the reliability of the simulation results directly depends on the accuracy of the input parameters. In the first part of this study, 13 March 1992 Erzincan (eastern Turkey) earthquake (Mw=6.6), which is recorded at only three strong ground motion stations, is simulated using the stochastic finite-fault method. The source and regional path parameters for this event are adopted from previously validated studies whereas the local site parameters are derived herein. In the second part of the paper, sensitivity of the simulation results with respect to small changes in selected input seismic parameters is investigated. The parameters for which sensitivities are computed include stress drop, crustal shear-wave quality factor and kappa operator. A change of 20% in stress drop value results in 14% change in PGA, whereas a 20% difference in the Q(0) value causes 17% change in PGA, and a 20% variation in kappa leads to 15% difference in PGA. Numerical experiments presented in this study prove that the ground motion simulations are prone to trade-off between the source, path and site filters. Hence, input models must be implemented carefully for reliable synthetic ground motions. (C) 2013 Published by Elsevier Ltd.