This paper presents an empirical modeling of a 2-Degree-of-Freedom (DoF) azimuth thruster using the signal compression method. The thruster has a gimbal mechanism with two servo motors and generates thrust in arbitrary directions. This mechanism can reduce the number of thrusters in an underwater robot and contribute to compact design. When an underwater robot is controlled with azimuth thrusters, the influence from the rotational motion of the thruster has to be considered, and a dynamic model of the azimuth thruster is needed. It is difficult to derive an analytical model because the system model depends on complicated fluid dynamics. In this study, empirical models of force and moment for rotational motion were derived for practical use through frequency analysis. A signal compression method can effectively extract the system model in the frequency domain from just the mechanically constrained frequency response. Experiments were carried out using a force/torque sensor that was connected to a cantilever in a water tank. The system model was analyzed with Bode plots, and the model coefficients were derived through curve fitting. The derived model was verified by a validation experiment.