We have developed a finite-element model of the 3D velocity structure of the Indo-Gangetic basin (IG basin) to quantify site amplifications due to seismic waves emanated from regional earthquakes. Estimating seismic wave amplifications is difficult in case of limited instrumentation, thus motivating us to propose a new simulation-based approach. The input required for the finite-element model include the spatial coordinates and the material properties (density, P-wave and S-wave velocities, Q factor) of the basin. Recent studies in the basin demarcate sediment layers of varying thickness, reaching down to a depth of 6 km and S-wave velocities ranging from 0.4-2.4 km/s (Srinivas et al., 2013). In the present study, our regional model has dimensions 900 x 900 x 80 km in x, y and z directions, discretized into 320 x 320 x 53 hexahedral elements. The top 6 km of the IG basin is divided into 8 different sediment layers with varying material properties. We use kinematic rupture models for the earthquake sources to simulate past as well as hypothetical future events. Two past earthquakes (Mw4.9, Delhi; Mw5.2, Chamoli) and two hypothetical earthquakes (Mw7.1; Mw8.5) are considered in our study. The rupture plane dimensions (L and W) and the slip distribution are estimated using the method of Mai and Beroza (2002). Based on focal-mechanism solutions and the depths of seismicity, we define the strike (580, 3090), the dip (650, 210), the rake (160, 770), and the depth of top edge of fault (5 km, 19 km) for the two large hypothetical earthquakes. Based on these parameters, the Centroid Moment Tensor (CMT) solution of the source is obtained. Ground motions are then simulated by solving the three-dimensional wave equation using the spectral element method (Komatitsch and Tromp, 1999). The key observations from our results are: 1) basin amplification factors for Peak Ground Velocity (PGV) are twice as high as Peak Ground Displacement (PGD) 2) PGV amplifications are as high as a factor of 6 for earthquakes occurring inside the basin, and a factor of 4 for Himalayan earthquakes (to the north of the study region) 3) The simulated shake maps of PGV and PGD show directivity. Based on the above observations, we conclude that it is important to include our model into low-frequency ground-motion estimation for seismic hazard analysis.