San Francisco’s earthquake history is defined by two events above all others: the catastrophic 1906 rupture along the San Andreas fault, which destroyed much of the city, and the 1989 Loma Prieta earthquake, which collapsed freeways and bridges during a live World Series broadcast. But behind those headline disasters lies a tectonic setting that makes the Bay Area one of the most seismically complex urban regions on Earth, with multiple parallel faults, soft sedimentary basins that amplify shaking, and recurrence intervals that ensure large earthquakes will keep coming.
The 1906 Earthquake and the Fire That Followed
On April 18, 1906, a rupture tore along roughly 470 kilometers of the San Andreas fault in northern California. The slip was enormous. Analysis of triangulation survey data from before and after the event shows that the fault moved as much as 8.6 meters near Shelter Cove and 7.5 meters at Tomales Bay, tapering to about 4.5 meters near Mount Tamalpais and 2.7 meters in the Santa Cruz Mountains near Loma Prieta.1Journal of Geophysical Research: Solid Earth. Resolution of fault slip along the 470‐km‐long rupture of the great 1906 San Francisco earthquake and its implications Those slip values are far larger than the 1.5 meters of surface offset that earlier surveys had measured, which means previous hazard assessments for parts of the fault had underestimated what 1906 actually released.
The earthquake itself, estimated at about magnitude 7.9, caused severe damage. But the fire that followed did far more. And the reason the fire burned unchecked for three days was simple: the water system failed. Multiple ruptures of the main pipeline trunk lines from the College Hill and University Mound reservoirs, caused by liquefaction-induced ground spreading and settlement, cut off more than 56 percent of the city’s total stored water supply to the Mission and downtown districts.2Earthquake Spectra. The 1906 San Francisco Earthquake and Fire—Enduring Lessons for Fire Protection and Water Supply The College Hill reservoir itself drained completely through two broken pipelines along Valencia Street, losing 53 million liters that could have supplied firefighters in the burning Mission District.
What remained was pitiful. Only the Clay Street Tank and the Lombard and Francisco Street reservoirs were within the fire zone and capable of delivering water directly. Their combined capacity was just 21 million liters, about 6 percent of total system storage. Even that small supply was further degraded by roughly 23,200 breaks in service connections, many caused by collapsing buildings. Fallen rubble blocked firefighters from reaching valves that could have shut off leaking mains.3Earthquake Spectra. The 1906 San Francisco Earthquake and Fire—Enduring Lessons for Fire Protection and Water Supply The lesson was brutal and enduring: in a city built on variable soils near active faults, a water system with trunk lines crossing liquefiable ground is a water system designed to fail at the worst possible moment.
The Fault System Under the Bay Area
People tend to think of “the San Andreas fault” as one crack in the ground, but the Bay Area’s seismic hazard comes from an entire system of parallel faults that together accommodate the motion between the Pacific plate and the North American plate. GPS and geodetic measurements put the total slip rate across this system at about 40 millimeters per year. That motion is split among several faults: the San Andreas itself carries roughly 17 mm/yr, while the Maacama fault takes about 14 mm/yr and the Bartlett Springs fault about 8 mm/yr.4Journal of Geophysical Research: Solid Earth. Kinematics of the Pacific‐North America Plate Boundary Zone, northern California Closer to the urban core, the Hayward fault on the east side of the bay also accumulates strain, complicated by the fact that it partially creeps between earthquakes, making it harder to determine exactly how much stored energy remains locked along it.5Journal of Geophysical Research: Solid Earth. Geodetic constraints on San Francisco Bay Area fault slip rates and potential seismogenic asperities on the partially creeping Hayward fault
This partitioning matters because it means no single fault carries all the hazard. A large earthquake on the San Andreas does not relieve the accumulated strain on the Hayward or Calaveras faults. Each fault has its own cycle, its own locked patches, and its own potential for a damaging rupture. For residents, that translates to a region where significant earthquakes can originate from several different directions, not just from the famous San Andreas trace running up the peninsula.
Why Some Neighborhoods Shake Harder Than Others
One of the most striking patterns in every Bay Area earthquake is the uneven distribution of damage. Buildings a few blocks apart can experience dramatically different shaking. The reason is geology: San Francisco and the surrounding bay shore sit on a patchwork of hard bedrock and soft sedimentary fill, and those soft sediments amplify seismic waves.
The Marina District, built partly on fill placed after the 1906 earthquake, is a textbook example. During the 1989 Loma Prieta earthquake, aftershock recordings in the Marina showed large amplification in ground motion and much longer duration of shaking compared to nearby rock sites, particularly at frequencies between 1 and 10 Hz. Some of that amplification comes from the simple difference in material stiffness between sediment and rock, but one-dimensional models consistently underpredict what actually happens between 1 and 3 Hz and cannot explain the prolonged shaking.6Bulletin of the Seismological Society of America. Modeling three-dimensional site response effects in the Marina District Basin, San Francisco, California Three-dimensional modeling reveals that the basin’s geometry itself plays a role: seismic waves entering the basin generate surface waves that bounce around within the soft sediment, focusing energy at certain spots and extending the shaking well beyond what the original seismic pulse delivered. The strongest recorded motions at basin sites came from these trapped surface waves, not from the initial arrival of the earthquake’s energy.
The same physics applies across the bay. Shallow sediments underlying San Francisco Bay, San Pablo Bay, Santa Clara Valley, and Livermore Valley all interact with seismic wave fields in ways that three-dimensional velocity models can now predict with reasonable accuracy.7Bulletin of the Seismological Society of America. Broadband Waveform Modeling of Moderate Earthquakes in the San Francisco Bay Area and Preliminary Assessment of the USGS3D Seismic Velocity Model At Treasure Island, for instance, a simple model of about 100 meters of stiff sediment sitting over weathered Franciscan rock reproduces the surface waves observed during recorded earthquakes.8Bulletin of the Seismological Society of America. The Effect of Shallow San Francisco Bay Sediments on Waveforms Recorded during the MW 4.6 Bolinas, California, Earthquake Understanding these basin effects is not just an academic exercise. It is what tells engineers which neighborhoods need stronger building codes and which infrastructure routes cross the most dangerous ground.
The 1989 Loma Prieta Earthquake
On October 17, 1989, a magnitude 6.9 earthquake struck in the Santa Cruz Mountains, about 95 kilometers from downtown San Francisco and Oakland. Despite the distance, the shaking in the Bay Area was severe in several places, and the damage pattern once again reflected the basin effects described above. Recordings at three rock sites near San Francisco showed brief, sharp shaking, suggesting an unusually short source duration for an earthquake of that size. But records at sedimentary sites in Oakland told a different story: strong amplification due to the local geology, with much more prolonged shaking.9Bulletin of the Seismological Society of America. The Loma Prieta earthquake, ground motion, and damage in Oakland, Treasure Island, and San Francisco
At Treasure Island, the seismic record showed coherent shear waves similar to the Oakland recordings until about 15 seconds in, when sudden steps in the acceleration signal marked the onset of liquefaction: the artificial fill literally turning to fluid under the cyclic shaking. The most visible catastrophe, though, was the collapse of the Cypress Street Viaduct, a double-deck freeway in Oakland that pancaked and killed 42 people. Ironically, the three closest recording stations formed a near-equilateral triangle around the collapse site and showed dominant shaking at a long period of roughly 1.5 seconds, far removed from the structure’s natural frequency of about 2.5 Hz for sideways motion.10Bulletin of the Seismological Society of America. The Loma Prieta earthquake, ground motion, and damage in Oakland, Treasure Island, and San Francisco The relationship between ground motion and structural failure turned out to be harder to pin down quantitatively than observers initially assumed, a reminder that connecting seismograms to collapse is still more complex than it looks.
Loma Prieta also reopened the 1906 water-supply conversation. While San Francisco’s system performed far better in 1989 than in 1906, the event underscored that pipeline breaks in liquefiable soils remain a vulnerability. The city has since invested heavily in its Auxiliary Water Supply System, a dedicated high-pressure network with cisterns and pump stations designed specifically for firefighting after an earthquake.
The Seventy-Year Quiet Period
One of the more curious patterns in Bay Area seismicity is the long calm that followed 1906. In the seven decades before the great earthquake, the region experienced 18 earthquakes of magnitude 6 or greater. In the 78 years after 1906, there were only four.11Journal of Geophysical Research: Solid Earth. Suppression of large earthquakes by stress shadows: A comparison of Coulomb and rate‐and‐state failure That is a dramatic drop, and it was not coincidence.
The explanation is a phenomenon called a stress shadow. When a huge earthquake releases strain along a fault, it also changes the stress conditions on nearby faults, often pushing them further from failure. The 1906 rupture was so large that it effectively turned down the seismic clock across the entire Bay Area for decades. Modeling shows that plausible combinations of friction and stress-loading parameters can account for the observed quiet period, assuming the pre-1906 earthquake rate would have continued without the great quake’s intervention.12Journal of Geophysical Research: Solid Earth. Suppression of large earthquakes by stress shadows: A comparison of Coulomb and rate‐and‐state failure
There was one notable exception. The 1911 Morgan Hill earthquake, a magnitude 6+ event on the Calaveras fault, struck just five years after 1906, right in the heart of the supposed stress shadow. The explanation lies in the Calaveras fault’s unusual behavior: the Morgan Hill segment has localized locked patches surrounded by sections that creep steadily. That steady creep means stress reloads on the locked patches extremely fast, fast enough to overcome the shadow imposed by 1906 within just a few years.13Geophysical Research Letters. A simple explanation for the occurrence of the 1911 Morgan Hill Earthquake in the stress shadow of the 1906 San Francisco Earthquake The Morgan Hill anomaly is actually a useful test case: it confirms the stress-shadow model by showing that the one fault that should have been able to break through the shadow is the one that did.
By the late 20th century, the shadow’s effect had largely worn off. The 1989 Loma Prieta earthquake was, in a sense, the Bay Area rejoining its normal seismic rhythm. Whether the current era represents a return to the pre-1906 rate of large earthquakes is still debated, but the protective blanket of the stress shadow is gone.
How Often the Big Ones Come Back
The recurrence interval for large San Andreas earthquakes near San Francisco depends on which segment of the fault you are asking about. On the North Coast segment, north of the Golden Gate, paleoseismic studies using offset creek channels and other geologic markers yield a long-term slip rate that, combined with the 4.9 to 5.5 meters of slip observed in 1906, gives a recurrence time for 1906-type earthquakes of roughly 221 years, plus or minus 40.14Geology. Late Holocene slip rate and recurrence of great earthquakes on the San Andreas fault in northern California That puts the next event in a statistical window that extends from the late 21st century into the 22nd, though statistical windows are not predictions. Earthquakes do not run on schedules.
Further south, in the Santa Cruz Mountains, trenching and tree-ring analysis at the Grizzly Flat site tell a different and somewhat surprising story. In a minimum span of about 340 years, only one large surface-faulting event occurred there: the 1906 rupture. The penultimate event dates to the mid-1600s, possibly as narrow a window as 1632 to 1659. Critically, there is no direct evidence in the trench record for the 1838 or 1865 earthquakes, which had long been proposed as occurring on this section of the fault.15Journal of Geophysical Research: Solid Earth. Paleoseismic investigations in the Santa Cruz mountains, California: Implications for recurrence of large‐magnitude earthquakes on the San Andreas Fault Earlier recurrence estimates for the Santa Cruz Mountains segment had been as short as 95 to 110 years. The trench data push that number considerably longer, suggesting this part of the fault may go centuries between major ruptures rather than a century or so.
The gap between these two pictures matters. If the North Coast segment has a roughly 220-year cycle and the peninsula segment goes longer, the two could rupture together in some events and separately in others, producing earthquakes of very different sizes. Paleoseismic records along strike-slip faults often show this kind of variability, where the same fault sometimes breaks in short segments and sometimes in long cascading ruptures.
Early Warning and the ShakeAlert System
Since 2019, the U.S. Geological Survey’s ShakeAlert system has been delivering earthquake early warnings to the West Coast through smartphone alerts and dedicated feeds. The system uses a network of seismometers to detect the initial, fast-traveling but less destructive compressional waves of an earthquake and issues alerts before the slower, more damaging shear waves arrive. Between October 2019 and September 2023, ShakeAlert generated 95 events with estimated magnitudes of 4.5 or above, the threshold set for public alerts.16Bulletin of the Seismological Society of America. Status and Performance of the ShakeAlert Earthquake Early Warning System: 2019–2023 That threshold was deliberately set below the level where damage typically begins, to give the system a margin for cases where it underestimates the earthquake’s true size.
For San Francisco, the practical value of ShakeAlert depends heavily on which fault produces the earthquake. A rupture on the Hayward fault, only about 20 kilometers from downtown, might give residents just a few seconds of warning. A repeat of the 1989 scenario, with a source 95 kilometers away in the Santa Cruz Mountains, could provide 15 to 20 seconds, enough time to drop under a desk or step away from windows. The system does not predict earthquakes; it races them, and the race is tighter the closer you are to the fault.
Retrofitting a Vulnerable Building Stock
San Francisco’s housing stock is famously old, and a large fraction of it consists of wood-frame buildings with “soft stories,” meaning the ground floor is open for garages or retail and lacks the bracing to resist lateral forces. These are the structures most likely to collapse or be rendered uninhabitable by moderate-to-strong shaking. San Francisco was among the first cities in California to mandate soft-story retrofits, requiring property owners to add steel frames or plywood shear walls to vulnerable ground floors.
Getting owners to actually complete retrofits, however, is harder than passing a law. Research into motivators and impediments for retrofit compliance in California finds that older, taller, and larger buildings tend to be the most structurally vulnerable but also carry the highest retrofit costs, creating a mismatch between urgency and willingness to invest.17PubMed Central. Motivators and impediments to seismic retrofit implementation for wood-frame soft-story buildings: A case study in California Commercial and residential owners respond to different incentives: residential owners are more influenced by safety concerns and insurance considerations, while commercial owners weigh the economic disruption of construction and the ability to pass costs to tenants. Mandatory programs with clear deadlines and enforcement have proven more effective than voluntary incentive-based approaches, but even in San Francisco, compliance timelines have been extended multiple times.
Beyond individual buildings, the region’s transit infrastructure faces its own challenges. BART, the commuter rail system that runs under the bay, includes tunnels that cross or run near active faults. Modeling of the Berkeley Hills tunnel’s response to fault displacement suggests that depending on how much the ground moves, the tunnels could sustain damage ranging from minor to significant, with repair times stretching from a few weeks to more than a year.18Elsevier. Response of BART’s Berkeley Hills tunnel to fault displacement and impact on system functionality For a region that depends on BART to move hundreds of thousands of commuters daily, even a weeks-long closure would have cascading effects on traffic, housing, and economic activity.
Living With Earthquakes That Have Not Happened Yet
San Francisco sits in an unusual psychological position among earthquake-prone cities. It has had two well-documented disasters in living or cultural memory, a tectonic setting that guarantees more, and a population that largely accepts the risk as background noise. Earthquake insurance uptake in California remains low relative to the actual hazard, partly because premiums are high and deductibles are steep, and partly because people tend to underestimate risks they have not personally experienced. For residents who arrived after 1989, no significant earthquake has struck the Bay Area in their lifetimes, and the stress shadow that once suppressed activity has been used up.
The practical reality is that the Bay Area’s faults are reloading. The roughly 40 mm/yr of plate motion across the fault system does not pause, and every year without an earthquake means more strain stored in the crust. The next large event could come from the San Andreas, the Hayward, the Calaveras, or one of the less famous faults that thread through the region. When it does, the outcome will depend heavily on the local geology underfoot, the integrity of the water system, the retrofit status of the building you happen to be inside, and how many seconds ShakeAlert can buy you before the shaking arrives.