We investigate the impact of considering a depth‐dependent attenuation model on source parameters assessed through a spectral decomposition. In particular, we evaluate the effect of considering the hypocentral depth as an additional variable for the attenuation model, using as the target the tendency of the average stress drop to increase with depth, as observed in recent studies. We analyze the Fourier spectra of S‐wave windows for about 1900 earthquakes with a magnitude above 2.5 recorded in the Ridgecrest region, southern California. Two different parameterizations of the attenuation term are implemented in the spectral decomposition, either as a function of the hypocentral distance alone or as a function of both epicentral distance and depth. The comparison of the spectral attenuation curves shows that, although the hypocentral model describes, on average, the range of values spanned by the attenuation curve for different depths, systematic differences with distance, depth, and frequency are observed. These differences are transferred to the source spectra and, in turn, to the source parameters extracted from the best‐fitting models. In particular, stress drops for events deeper than 7 km are, on average, almost double even when depth is introduced explicitly in the attenuation model. The increase of stress drop with depth is confirmed also after accounting for the increase of the shear velocity with depth, which absorbs about 30%–40% of the total increase. Moreover, a qualitative comparison with a model for the gradient of the effective normal stress confirms the reliability of the observed trend. Finally, the coherent spatial patterns shown by a simplified 2D tomographic representation of the spectral residuals highlights the impact on ground‐shaking variability of the lateral variability of the crustal attenuation properties in the region.