The time‐averaged shear‐wave velocity of up to 30 m (VS30), a vital parameter in building codes, is widely used in site classification and plays an important role in the analysis of site effects. Recently, the log–linear correlation (Boore, 2004) has become more prevalent to estimate VS30 of velocity profiles that do not extend to 30 m. In this study, we make a full analysis on the linear correlation with three publicly available data sets of shear‐wave velocity profiles. A conditional independence property is derived from the partial correlation analysis of the data sets. Using this property, a new VS30 estimation model is proposed. Comparison of models based on the setup of training and testing data sets suggests that the new model achieves improved performance on the VS30 estimation. The impact of errors of estimation models on the prediction of ground‐motion parameters is also evaluated, and the overall error of the predicted parameter by the new model is the least, especially at long periods.