Traces of many normal faults form an array of closely spaced overstepping segments. In three dimensions, fault segments may either be unconnected or link vertically or laterally into a single continuous fault surface. The slip distribution along segmented faults is complex and asymmetric, and the point of maximum slip generally is not located at the center of a segment. In relay zones between segments, slip gradients may be gentler or steeper, depending on the spatial fault arrangement. Branch points are characterized by steep slip gradients. One explanation for these observations is mechanical interaction between neighboring faults which occurs through local perturbation of the stress field. Three-dimensional (3-D) boundary element models show that the degree of fault interaction and hence the degree of asymmetry in the slip distribution increases with increasing fault overlap and downdip fault height and with decreasing fault spacing and Poisson's ratio. Interaction is strongest for faults with uniform shear strength and decreases if there exists a zone of greater shear strength near the tip line. This analysis provides a mechanical rationale for more frequent occurrence of overlapping segments relative to underlapping segments and for the limited range of the ratio between segment overlap and spacing along natural faults. Echelon segment configurations promote interaction, maximize the capacity to accommodate slip, and do not necessarily require a strike-slip movement component. Model idealizations of some outcropping fault arrays and of branching/merging faults capture a wide variety of common field observations. Consistent, mechanically based 3-D normal fault models can be obtained by combining different types of field data such as fault slip-to-length ratios, location of maximum slip, segment overlap-to-spacing ratios, and footwall uplift/hanging wall subsidence. By capitalizing on these data one can understand the mechanics of faulting, constrain the boundary conditions that govern the formation and growth of faults, and provide a rationale for interpreting normal faults in seismic surveys.¿ 1997 American Geophysical Union