The potential rupture length of an active fault is primary input, and an uncertainty, in source characterization for seismic hazard analysis. The past several decades have seen source models in which faults are divided into potential rupture segments based on fault‐specific behavioral and geometric observations. These models (prescribed segmentation) produced single and multisegment ruptures for use in earthquake probability and regional ground‐motion estimates. The 2013 Uniform California Earthquake Rupture Forecast, Version 3 (UCERF3), relaxed segmentation through modeling of fault‐to‐fault jumps that result in ruptures along the San Andreas fault system as long as 1200 km. Principle controls of UCERF3 rupture length are the stepover distance between different faults or fault sections, and their orientation relative to modeled coulomb stress changes. Observations of surface faulting worldwide since the mid‐1800s in shallow continental crust show that only 30 of the reported ruptures, about 12% of the reported examples, have been longer than 100 km; only 6 have been longer than 300 km and none has exceeded 500 km. A review of characteristics of long ruptures shows that, in general, they occur along essentially continuous traces with limited stepping or branching from one fault to another as opposed to rupture propagation on linkages of different faults with highly variable physical characteristics. These commonly occupy from 10% to 70% of the total host fault length. Neither segmented nor relaxed models of rupture length have been tested by the occurrence of actual ruptures. There is no reason why segmentation models incorporating timing, dynamic, and rheological factors cannot provide reasonable approximations of future rupture length. For many faults of concern worldwide, this may be the most effective approach.