Strong ground motions recorded during the 2008 Wenchuan, China, earthquake (M_w 7.9) have been simulated using the stochastic finite-fault method proposed by Beresnev and Atkinson (1997, 1998b). The simulations were made for two source models. Both models are based on the fault geometry that was proposed by Koketsu et al. (2008) through inversion of teleseismic body wave data. The slip distribution obtained by this inversion was used for the first source model, while a random slip distribution was used for the second source model. The performance of each source model is quantified by calculating the bias and standard deviation of response spectra predicted by each model. For the first source model, the results show overall agreement between the simulated and observed response spectra in a period range of 0.05–1 s, as well as 4–10 s, but the model overpredicts ground motions in a period range of 1–4 s. For the second source model, the model is biased over a slightly wider period range at longer periods. The performance of the stochastic model to predict observed ground motions is also compared with several empirical ground-motion models by means of statistical tools.