The basic assumption in the prediction of peak ground acceleration (PGA), peak ground velocity (PGV), and pseudoresponse spectral values by the random vibration theory (RVT) method is that the ground motion process is a bandlimited Gaussian random process (BGRP). However, for the estimation of pseudoresponse spectral values, the process is the output of a single-degree-of-freedom (SDOF) system subjected to the input of a BGRP. The output process is a narrow-band random process because a SDOF system acts as a narrow-bandpass filter. The property of a narrow-band process is significantly different from that of a bandlimited process. There is an obvious difference in the estimations of the pseudoresponse spectral values based on bandlimited or narrow-band process, especially in the lower frequency part. In this study, we propose an empirical method to improve the estimation of the pseudoresponse spectral values by the RVT based on the consideration of properties of a narrow-band Gaussian stationary process. Comparisons of our results with those of previous research studies and the time domain simulation (TDS) shows that our empirical approach improves the estimation of pseudoresponse spectral values in the long period range.