We develop new empirical scaling laws for rupture width W, rupture length L, rupture area A, and average slip D, based on a large database of rupture models. The database incorporates recent earthquake source models in a wide magnitude range (M 5.4–9.2) and events of various faulting styles. We apply general orthogonal regression, instead of ordinary least-squares regression, to account for measurement errors of all variables and to obtain mutually self-consistent relationships. We observe that L grows more rapidly with M compared to W. The fault-aspect ratio (L/W) tends to increase with fault dip, which generally increases from reverse-faulting, to normal-faulting, to strike-slip events. At the same time, subduction-inter-face earthquakes have significantly higher W (hence a larger rupture area A) compared to other faulting regimes. For strike-slip events, the growth of W with M is strongly inhibited, whereas the scaling of L agrees with the L-model behavior (D correlated with L). However, at a regional scale for which seismogenic depth is essentially fixed, the scaling behavior corresponds to the W model (D not correlated with L). Self-similar scaling behavior with M − log A is observed to be consistent for all the cases, except for normal-faulting events. Interestingly, the ratio D/W (a proxy for average stress drop) tends to increase with M, except for shallow crustal reverse-faulting events, suggesting the possibility of scale-dependent stress drop. The observed variations in source-scaling properties for different faulting regimes can be interpreted in terms of geological and seismological factors. We find substantial differences between our new scaling relationships and those of previous studies. Therefore, our study provides critical updates on source-scaling relations needed in seismic–tsunami-hazard analysis and engineering applications.