This article develops a method to generate ground motion time histories that maximize the response of a given linearly elastic structure. The root mean square (RMS) level of the input power spectral density (PSD) is used as a strong motion parameter. It is related to seismological data that is readily available. An empirical relation to estimate RMS value of the PSD from peak ground acceleration, magnitude, rupture distance, and shear wave velocity is derived from world-wide strong motion data. The ground motion is obtained by solving the inverse problem such that the structural response is maximized under the constraint of fixed value of RMS level of the input PSD enforced using a Lagrange multiplier. The proposed methodology is illustrated for a single-degree of freedom system, a six storey building and an earthen dam. It is shown that the critical PSD obtained in all the cases is a narrow band process resulting in stochastic resonance and not a Dirac-delta function with the entire energy of the system concentrated at its natural frequency. Moreover, the critical excitation samples generated using this critical PSD resembles actual earthquake acceleration time histories. © A. S. Elnashai.