Current research on ground‐motion models (also known as ground‐motion prediction equations [GMPEs]) and their uncertainties focus on the separate contributions of source, path, and site to both median values and their variability. Implicit here is the assumption that the event term, path term, and site term reflect only properties of the source, path, and site, respectively. Events with larger stress drop generate more high‐frequency energy, and thus more ground motion. Therefore, the correlation of high‐frequency (i.e., peak ground acceleration [PGA] or peak ground velocity [PGV]) event terms in GMPEs with stress drop is taken to be genuine. However, PGA and PGV ground‐motion observations of the 2014 M 6.0 South Napa, California, earthquake clearly violate these assumptions. For this earthquake, high‐frequency ground‐motion residuals of recorded ground motion with respect to Next Generation Attenuation‐West2 Project (NGA‐West2) ground‐motion models show a dependence on distance, biasing the calculation of the event term by incorrectly mapping a regional attenuation effect into it. We examine the trade‐off between source and path effects for the South Napa earthquake and a well‐recorded California subset of the NGA‐West2 data. We fit near‐source (i.e., within 20 or 50 km) event terms and remaining differential geometrical spreading and anelastic attenuation terms in comparison to a simultaneous inversion for the source and path terms. This South Napa instance highlights one situation for which the high‐frequency event term can be interpreted as relative stress drop only when the distance dependence of the ground motions does not bias the residuals.