An application of a linear multivariate Bayesian regression model, described in a companion article, to obtain a ground-motion prediction equation (GMPE) using a set of actual ground-motion records and a realistic functional form is presented. Based on seismological grounds and on an adopted functional form, we include a sound discussion about how the prior information required for the model can be defined. For the regression analyses we use two subsets of ground-motion records from the Next Generation of Ground-Motion Attenuation Models (NGA) database. We compare the results obtained with the Bayesian model with those obtained through the one-stage maximum-likelihood and the constrained maximum-likelihood methods. The advantages of the Bayesian approach over traditional regression techniques are discussed.