在實務上土層受震反應分析多採用一維土層模式，然而於軟弱土層所進行的局部地盤改良會使得土層不再符合一維土層模式的假設而成為三維之問題。 本研究採用三維有效應力非線性有限元素動力分析程式，分析局部地盤改良土層於不同人工地震作用下之受震反應，並與使用一維和二維土層模式所求得之的地表震動反應譜與座落在該改良區內之10層建築物的最大層間變位角進行比較。 分析局部改良地盤的土層受震反應時，使用一維土層模式會嚴重低估土層之反應，而二維及三維土層模式所獲致的反應譜最大值有差異，但所對應之週期卻相當接近，因此建議分析局部改良地盤的土層受震反應時應使用二維或三維模式。當改良區塊邊長大於400公尺以上時，可採用一維土層模式分析。在建築的最大層間變位角方面，相對於二維及三維土層模式而言，使用一維土層模式的地表地震動做為輸入的結果較不保守。此外，對於建築物之最大層間變位角即使依規範由本研究的十組結果中以三種不同的組合方式，每次任取七組的平均結果仍有差距，可見此作法亦無法有效減低種數的影響。 In practice, the 1D model is frequently used to analyze the seismic responses of a soil stratum. When the ground improvement is performed over a finite region of a soft soil stratum, the 1D assumption of the ground is violated and the problem becomes a 3D one. In this study, a 3D nonlinear finite element model is used to compute the surface ground motions of an improved soil zone. The results are compared with the previously obtained results using 1D and 2D ground models. Their effects on the seismic response analyses of a ten-story building are also studied. For an improved soil zone, the seismic responses computed using 1D model will be underestimated, while the 2D and the 3D ground models show similar acceleration spectra, with 2D model results showing higher values. Thus, for such a case, the surface ground motion must be obtained using the 2D or the 3D ground model if the design ground motion is specified at the engineering bedrock. When the width of the improved zone is larger than 400 meters, the 1D model can be adopted to simulate its seismic responses. For the dynamic responses of a building located on top of an improved soil zone, if the input ground motion is computed using the 1D ground model, then the response will be underestimated; thus, for such a case, the surface ground motion must be obtained using motions obtained by the 2D or the 3D ground models. will be better. The seed numbers for generating the artificial earthquakes have significantly effects on the seismic responses of ground and the maximum drift ratio of the building. Although it has been proposed that the average value of the responses from at least seven artificial earthquakes be used, some differences still can be observed for different combinations from ten responses of artificial earthquakes.