The roll response of the damaged ship in waves is often calculated by establishing the roll motion equation in the study of the damaged ship stability. The accuracy of the roll hydrodynamic coefficients in the roll motion equation needs to be ensured for the accuracy of the calculation results. The roll hydrodynamic coefficients of the damaged ship vary with the roll amplitude and the roll frequency. By analyzing the characteristic of the roll hydrodynamic coefficients of the damaged ship, a fast prediction method is proposed to predict the hydrodynamic coefficients for damaged ships under different roll amplitudes and frequencies. The forced roll oscillation of a damaged frigate DTMB-5415 under different roll amplitudes and frequencies is simulated by using the computational fluid dynamics (CFD). The roll moments are regressed to obtain the added moments of inertia and damping coefficients. The linear interpolation method and the cubic spline interpolation method are adopted to establish the hydrodynamic coefficient database by taking the roll hydrodynamic coefficients calculated by CFD as sample points. The influence of the two interpolation methods and the interval of the sample points on the calculation accuracy are analyzed by comparing the predicted hydrodynamic coefficients with those obtained by CFD method. The damping coefficients increase monotonically with the increase of the roll amplitude.However, the trend of the added moment of inertia varying with the amplitude shows obvious nonlinearity. The hydrodynamic database established by the two interpolation methods can basically reflect the characteristics of the roll hydrodynamic coefficient varying with the roll amplitude. The roll hydrodynamic database is recommended to be established by the cubic spline interpolation method with the sample point interval of 5 degree for the roll amplitude with comprehensive consideration of the influence of the hydrodynamic coefficient characteristics and the interval of sample points on the accuracy of the hydrodynamic database. By comparison with the CFD results, the average error and maximum error are about 6% and 11%, respectively. It proves that the proposed interpolation method can efficiently predict the roll hydrodynamic coefficients of the damaged ship.