In this paper, a trial-error based method is presented to calibrate added mass models through numerical iterations minimizing the difference between the measured frequency data and its numerical simulation result for a dam floodgate. Earthquake analysis of the real floodgate for which the on-site hammering vibration test is performed show that the classical Westergaard added mass model gives relatively larger values in the maximum earthquake force and the maximum total displacement than the present added mass model, based on the calibration of on-site measurement data.