Near-fault ground motions often possess distinct characteristics, which can have strong influence on structural response. These include the rupture directivity effect in the fault-normal direction and the fling step in the fault-parallel direction. A site in the near-field region of the fault may experience forward directivity when the fault rupture propagates towards the site with a velocity almost equal to the shear-wave velocity of the ground medium. The resulting ground motion typically exhibits a two-sided, long-period, large-amplitude velocity pulse in the fault-normal direction. Backward directivity occurs when the fault rupture propagates away from the site. The resulting ground motion tends to be of low intensity and long duration. The fling step is caused by the permanent displacement of the fault and is usually characterized by a one-sided velocity pulse in the fault-parallel direction. Due to scarcity of recorded near-fault ground motions, there is interest in developing synthetic ground motions for near-fault sites, which can be used in performance-based earthquake engineering in addition to or in place of recorded motions. It is crucial that such synthetic motions be realistic and have characteristics that are consistent with recorded near-fault ground motions. Furthermore, from a practical standpoint, it is most useful if the synthetic motions are generated in terms of information that is normally available to the design engineer. This information typically includes the type of faulting, the earthquake magnitude, the position of the site relative to the potential fault rupture, and the shear-wave velocity of the soil at the site. In this report we develop a parameterized stochastic model of near-fault ground motion in the strike-normal direction. Not all near-fault ground motions contain a forward directivity pulse, and our model is developed to account for the pulselike and non-pulselike cases. By fitting the model to a database of near-fault ground motions, we develop predictive equations for the model parameters in terms of the earthquake source and site characteristics mentioned above. Using these predictive equations, for a given set of earthquake and site characteristics, we generate sets of model parameters and use them to generate an ensemble of synthetic near-fault ground motions. The resulting synthetic motions have the same statistical characteristics as the motions in the database, including the variability for the given set of earthquake and site characteristics. For illustration of the methodology, a set of synthetic motions for specified earthquake source and site characteristics are generated and their characteristics are compared with those of recorded motions.