To estimate the fault dimension of an earthquake in real time, we present a methodology to classify seismic records into near-source or far-source records. Characteristics of ground motion, such as peak ground acceleration, have a strong correlation with the distance from a fault rupture for large earthquakes. This study analyzes peak ground motions and finds the function that best classifies near-source and far-source records based on these parameters. We perform (1) Fisher’s linear discriminant analysis and two different Bayesian methods to find the coefficients of the linear discriminant function and (2) Bayesian model class selection to find the best combination of the peak ground-motion parameters. Bayesian model class selection shows that the combination of vertical acceleration and horizontal velocity produces the best performance for the classification. The linear discriminant function produced by the three methods classifies near-source and far-source data, and in addition, the Bayesian methods give the probability for a station to be near-source, based on the ground-motion measurements. This discriminant function is useful to estimate the fault rupture dimension in real time, especially for large earthquakes.