Earthquake ground‐motion prediction models usually define site conditions based on the time‐averaged shear‐wave velocity in the upper 30 m ( V S 30). Proxy‐based estimations of V S 30 are commonly used, if velocity measurements are not available. We compile a soil‐profile database for the Beijing plain area (China), using data from research documents and technical reports. The database contains 479 soil profiles, 463 of which have depths greater than 30 m. We develop regional relationships for the Beijing plain area for extrapolating the time‐averaged shear‐wave velocity to a given depth less than 30 m to V S 30, and then compare the performance of available models. We find that the second‐order polynomial model (Boore et al. , 2011), based on data from Japan, provides an overprediction, whereas the linear model (Boore, 2004) calibrated on data from California underestimates V S 30. We develop relationships for estimating V S 30 based on proxies such as ground slope gradients from radar‐derived digital elevation models (DEMs) and surface geology at different scales. We find that local V S 30 data in the Beijing plain are generally lower than existing 30 arcsec gradient‐based global models. Regression results show a modest correlation between V S 30 and topographic ground slope for several DEM resolutions (3, 15, 30, and 60 arcsec). Geology‐based proxies are more effective than ground slope for V S 30 estimation in the analyzed area. We propose a bilinear model based on geologic ages and depositional environments for estimating V S 30, which shows a statistically significant trend for application in the Beijing plain area. Online Material: Figures showing topographic ground slopes and correlations of V S 30 with topographic slope from digital elevation model (DEM) data and a table summarizing data from the 463 boreholes.