We compared response spectra from the Mw 7.9 2008 Wenchuan earthquake with five modern ground-motion prediction equations (GMPEs). Ninety-three strong-motion records within 300 km of the fault plane were selected for comparison with the GMPE models of Zhao, Zhang et al. (2006), Abrahamson and Silva (2008), Boore and Atkinson (2008), Campbell and Bozorgnia (2008) and Chiou and Youngs (2008) for spectral periods up to 5.0 s. The site class of the recording stations used for the Zhao, Zhang et al. (2006) model was inferred from response spectral ratios of the horizontal and vertical components (H/V) computed from the strong-motion records in moving and overlapping time windows. The average shear-wave velocity of the top 30 m (VS30) was only available for two stations. VS30 was extrapolated from the average of the top 20 m (VS20) when possible and inferred from the H/V response spectral ratios when necessary. The average predictions of all models were acceptable. The Zhao, Zhang et al. (2006) model gave the best predictions for peak ground acceleration and short spectral periods, especially up to 100 km of the source distance. All Next Generation Attenuation (NGA) models predicted the recorded spectra very well for periods of 0.5–1.0 s and at 5.0 s. The Chiou and Youngs (2008) model gave the best overall predictions. The standard deviations of all attenuation models were similar at a 5% significance level. However, differences between spectra estimated by various NGA models were statistically and practically significant, with the largest difference between the average predictions being nearly a factor of 1.4 at the 0.1-s period and 2.3 at the 5.0-s period for data within a source distance of 100 km. Although one earthquake did not produce median ground motions that the GMPEs are designed to predict, such a large difference represents a challenge for empirical models when estimating spectra from very large crustal earthquakes.