The possible amplification of seismic waves in soil deposits is crucial for the seismic design of buildings and geotechnical systems. The most common approaches for the numerical simulation of seismic site response are the equivalent linear (EQL) and the nonlinear (NL). Even though their advantages and limitations have been investigated in several studies, the relative field of applicability is still under debate.
This study tested both methods over a wide population of soil models, which were subjected to a set of acceleration time histories recorded from strong earthquakes. A thorough comparison of the results of the EQL and the NL approaches was carried out, to identify the conditions in which the relative differences are significant. This assessment allowed for the definition of simplified criteria to predict when the two schemes are or are not compatible for large expected shaking levels. The proposed criteria are based on simple and intuitive parameters describing the soil deposit and the ground‐motion parameters, which can be predicted straightforwardly. Therefore, this study provides a scheme for the choice between the EQL and the NL approaches that can be used even at the preliminary design stages. It appears that the EQL approach provides reliable amplification estimates in soil deposits with thickness up to 30 m, except for very deformable soils, but this depth range may be extended at long vibration periods. This result reveals a good level of reliability of the EQL approach for various soil conditions encountered in common applications, even for high‐intensity shaking.