THE Northridge earthquake of 17 January 1994(1) was the latest in a series of very damaging, thrust-fault-generated earthquakes to strike California, following the San Fernando(2) 1971, Coalinga(3) 1983, and Whittler Narrows(4,5) 1987 events. Like the last two of these, the Northridge event occurred along a fault that did not reach the surface and which had not been detected by traditional seismic-hazard methods(6,7). Balanced cross-sections(8,9), which flatten and remove the crustal deformation, can be used to identify and quantify the seismic hazard posed by thrust faults. Here we present a balanced cross-section through the Northridge portion of the Transverse Ranges fold-and-thrust belt(10), which shows that the earthquake occurred on what we call the Pico thrust. A cross-section of this sort constructed before the earthquake would have revealed the fault, although it would not have predicted the earthquake. Cross-sectional modelling of the Pico thrust yields an average slip rate of 1.4-1.7 mm yr(-1) and a recurrence interval of Northridge-sized (M(w) 6.7) earthquakes every 1,500-1,800 years. We show that the Pico thrust is the back thrust to the 170-km Elysian Park thrust(3,4) which underlies some of the most densely urbanized portions of the Los Angeles basin.