The random‐vibration theory (RVT) approach to equivalent‐linear site‐response analysis is often used to simulate site amplification, particularly when large numbers of simulations are required for incorporation into probabilistic seismic‐hazard analysis. The fact that RVT site‐response analysis does not require the specification of input‐time series makes it an attractive alternative to other site‐response methods. However, some studies have indicated that the site amplification predicted by RVT site‐response analysis systematically differs from that predicted by time‐series approaches. This study confirms that RVT site‐response analysis predicts site amplification at the natural site frequencies as much as 20%–50% larger than time‐series analysis, with the largest overprediction occurring for sites with smaller natural frequencies and sites underlain by hard rock. The overprediction is caused by an increase in duration generated by the site response, which is not taken into account in the RVT calculation. Correcting for this change in duration brings the RVT results within 20% of the time‐series results. A similar duration effect is observed for the RVT shear‐strain calculation used to estimate the equivalent‐linear strain‐compatible soil properties. An alternative to applying a duration correction to improve the agreement between RVT and time‐series analysis is the modeling of shear‐wave velocity variability. It is shown that introducing shear‐wave velocity variability through Monte Carlo simulation brings the RVT results consistently within ±20% of the time‐series results.