The dynamic response of nuclear power plants and other critical infrastructure under seismic conditions is a major issue, and the earthquake-resistant capacity design of such facilities and their interior equipment should be based on prediction of dynamic response. Therefore, field geological investigations and corresponding site response analyses facilitate the evaluation of seismic site effects for the entire construction process, including site selection and detailed design. During site response analysis, soil parameters are usually obtained from laboratory tests using specimens retrieved from boreholes or from empirical equations. Conventional site response analysis software assumes horizontal layering in a 1D soil model, though the actual conditions of underground stratification are far more complicated. Some research efforts have undertaken 3D site response and soil-structure interaction analysis using LS-DYNA; however, the influence of different boundary conditions in these studies has been rarely discussed. The research reported on in this paper utilized LS-DYNA to perform nonlinear time-domain site response analysis and evaluated the difference between 1D and 3D model results. The results of both 1D and 3D site response analyses conducted using LS-DYNA were comparable to those of traditional approaches using a constrained boundary. The 3D models that adopt a non-reflective boundary condition can be affected by energy attenuation near the model boundary, causing a significant difference in the results. We suggest that the constructed model should have an area of at least 600 m × 600 m (plan view) to ensure accuracy in a 200 m × 200 m area in the center of the 3D simulation model using non-reflecting boundary conditions. Furthermore, four soil profile from real sites have been verified that the suggestion is also applicable in multilayered scenario.