A new efficient approach for generating spectrum-compatible seismic accelerograms is proposed. It is based on the Hilbert-Huang Transform (HHT); one natural seismic accelerogram is decomposed into frequency and amplitude components. The components are appropriately modified to synthesize the artificial seismic accelerogram that appears to have compatible acceleration spectrum with the natural seismic accelerogram. The HHT is an adaptive signal processing technique for analyzing nonlinear and non-stationary data such as seismic accelerograms. With HHT a seismic accelerogram is decomposed into a finite and small set of components. These components have well defined instantaneous frequencies, estimated by the first derivative of the phase of the analytic signal. The method is tested using twenty natural seismic records and a comparison with two established methodologies is provided.