The time‐averaged shear‐wave velocity to a depth of 30 m (VS30) is widely used in various geotechnical and seismic issues such as ground‐motion prediction equations, determination of site class in building codes, and evaluation of site effects. However, in many cases the velocity is not measured as deep as 30 m. Therefore, obtaining a correlation between VS30 and velocities at depths lower than 30 m (VSz) seems to be necessary. The present article proposes a rigorous model selection approach to find the best model for estimation of VS30 by VSz based on 257 Iranian profiles. The candidate models include the Boore et al. (2011) and Chandler et al. (2005) relations, as well as two types of power‐form correlation presented by the authors. The presented approach includes hypothesis testing based on goodness‐of‐fit measures of normalized residuals, the likelihood method of Scherbaum et al. (2004), and finally the log‐likelihood (LLH) criterion based on the information theory. The results show that the best model obtained by each of these methods is not necessarily the same. Furthermore, it is observed that while hypothesis tests are somehow inconsistent and noninformative, the LLH criterion presents a coherent method for ranking various candidate models and selecting the best one.