Brittle deformation in the upper crust is thought to occur primarily via faulting. The fault length-frequency distribution determines how much deformation is accommodated by numerous small faults vs a few large ones. To evaluate the amount of deformation due to small faults, we analyze the fault length distribution using high-quality fault maps spanning a wide range of spatial scales from a laboratory sample to an outcrop to a tectonic domain. We find that the cumulative fault length distribution is well approximated by a power law with a negative exponent close to 2. It follows that faulting is a self-similar process, and a substantial fraction of tectonic strain can be accommodated by faults that don’t cut through the entire brittle layer, consistent with inferences of “hidden strain’ from natural and laboratory observations. A continued accumulation of tectonic strain may eventually result in a transition from self-similar fault networks to localized mature faults.