A conventional lumped-mass stick model is based on the tributary area method to determine the masses lumped at each node and used in earthquake engineering due to its simplicity in the modeling of structures. However the natural frequencies of the conventional model are normally not identical to those of the actual structure. To solve this problem, recently an updated lumped-mass stick model is developed to provide the natural frequencies identical to actual structure. The present study is to investigate the seismic response accuracy of the updated lumped-mass stick model, comparing with the response results of the shaking table test. For the test, a small size four-story steel frame structure is prepared and tested on shaking table applying five earthquake ground motions. From the comparison with shaking table test results, the updated model shows an average error of 3.65% in the peak displacement response and 9.68% in the peak acceleration response. On the other hand, the conventional model shows an average error of 5.15% and 27.41% for each response.