The earthquake ground-motion prediction equations developed for the Next Generation Attenuation of Ground Motions (NGA–West) project in 2008 have established a new baseline for the estimation of ground-motion parameters, such as peak ground acceleration, peak ground velocity, and spectral acceleration, for shallow crustal earthquakes in active tectonic regions. We perform statistical goodness-of-fit analyses to quantitatively compare the predictive capabilities of the NGA models and their predecessors, using several testing subsets of the master database used in model development. In addition, we perform a blind comparison test using 1060 ground-motion records from seven recent earthquakes recorded in California: the 2003 M 6.5 San Simeon event, 2004 M 6.0 Parkfield event, 2005 M 5.2 Anza event, 2007 M 5.4 Alum Rock event, 2008 M 5.4 Chino Hills event, 2010 M 7.2 Baja event, and 2010 M 5.7 Ocotillo event. We assess how modeling decisions regarding the regression dataset, functional forms, input parameters, and model complexity influence the models’ predictive capabilities. By comparing the performance of each model, we discuss various ground-motion modeling strategies and offer recommendations for model development. We find that increased model complexity does not necessarily lead to increased prediction accuracy, that the inclusion of aftershocks in regression datasets may result in decreased predictive capabilities for mainshocks, and that the use of measured site characteristics leads to greatly improved ground-motion predictions. A model validation framework is introduced to assess the prediction accuracy of ground-motion prediction equations and to aid in their future development.