A database method is proposed to predict the roll response of a damaged ship upon quasi-steady flooding in regular beam waves. To construct the hydrodynamic database, the hydrodynamic loads acting on the damaged ship are decomposed into wave excitation moment and motion-induced moment, which are calculated based on computational fluid dynamics (CFD). The wave excitation moments for different wave frequencies are formulated using the high-order trigonometric function. The ship hydrodynamic coefficients (i.e., added moment of inertia and damping coefficient) for different roll frequencies and amplitudes are regressed from the motion-induced moment and then used as samples for database construction by cubic spline interpolation. The Runge-Kutta method is employed to solve the ship's roll motion equation, in which the hydrodynamic coefficients are updated according to the roll amplitude in real time. To test the performance of the present method, the roll motion responses of a damaged frigate DTMB-5415 are computed. The results obtained by the database method are in good agreement with the CFD results and experimental data. The characteristics of motion-induced moment, roll hydrodynamic coefficients and wave excitation moment are analyzed. In addition, the influence of various factors (e.g., sensitivity of mesh and time step, order of regression method, interval of sample points and type of load decomposition model) on the computation accuracy are also discussed. Compared to directly performing CFD simulation of damaged ship motion in waves, the proposed database method can predict the ship's roll response with high efficiency and acceptable accuracy.