On March 18, 1925, a tornado track crossed Missouri, Illinois and Indiana and killed roughly 695 people, which remains the highest toll from a single tornado in United States history. No warning reached anyone ahead of it, in part because the Weather Bureau was not permitted to use the word tornado in a forecast. This piece covers what is documented about that afternoon, why the 695 figure gets confused with the larger toll for the whole day, how warning doctrine actually developed in the decades that followed, and why a long-track tornado is a different planning problem than a tornado that hits one town.

What happened on March 18, 1925

The storm is generally described as beginning in the early afternoon in Reynolds County, Missouri, near Ellington, moving east-northeast across the Mississippi River into southern Illinois, and ending in the late afternoon in southwestern Indiana past Princeton. Standard references have long given the path as roughly 219 miles covered in about three and a half hours, which works out to an average forward speed near 60 miles per hour, with some accounts putting the movement faster than that over portions of the track. In Illinois it struck Gorham, Murphysboro, De Soto, Hurst, Bush, West Frankfort and Parrish in succession, and in Indiana it destroyed Griffin and tore through the edge of Princeton.

Accounts from survivors collected at the time and afterward describe something that did not look like the tornado people expected, because the circulation was wrapped in rain and dust and in places was described as a rolling fog or a boiling cloud at ground level rather than a visible funnel. That matters operationally even now, since rain-wrapped tornadoes defeat visual confirmation, and in 1925 visual confirmation by a person standing outside was the only detection method available to anyone. Forward speed compounded the problem, because a storm moving at highway speed gives a bystander who does see something almost no time to interpret it.

Modern references rate the tornado F5, and that rating is retrospective. Theodore Fujita did not publish his damage scale until 1971, so every pre-1971 rating in the record is a reconstruction from photographs, newspaper descriptions, structural damage accounts and later field review, applied by researchers decades after the fact. The Enhanced Fujita scale replaced the original for operational use in the United States on February 1, 2007, which adds another translation layer between the 1925 damage and the way a damage survey would be written today.

The death toll, and the number it gets confused with

The figure of 695 deaths is the one published by the National Weather Service for the Tri-State track and the one carried in Thomas Grazulis’s reference work on significant tornadoes, and it is the basis for the statement that this is the deadliest tornado in US history. Within that total, Murphysboro, Illinois accounts for roughly 234 deaths, which is the highest toll any single American community has suffered from one tornado. De Soto lost around 69 people, and the school there collapsed with children inside, with the number of children killed at that school commonly reported as 33. West Frankfort is usually given at around 148 dead, a figure shaped by the fact that it was a coal town where a large share of the men were underground at the time and came up to find their neighborhoods gone. Injuries across the track are commonly cited at just over two thousand.

The number that gets substituted for 695 is 747, and the two are not measuring the same thing. March 18, 1925 produced a broader outbreak, with additional tornadoes reported in Kentucky, Tennessee, Alabama, Kansas and elsewhere in Indiana, and compilations of that whole day’s fatalities run higher than the single-track figure. When you see 747 attached to the Tri-State Tornado itself, you are usually looking at the outbreak total misapplied to one storm. When you see 695, you are looking at the single-track figure. Both numbers appear in reputable places, they disagree because they count different events, and the honest way to write either one is to say which it is.

There is a second and more interesting complication, which is whether 695 should be attributed to one tornado at all. A 2013 re-analysis published in the Electronic Journal of Severe Storms Meteorology by Robert Johns, Donald Burgess, Charles Doswell and colleagues went back through newspaper records, survey material and field review of the path and argued that the traditional continuous track needs revision, with portions at the ends of the long-accepted path likely belonging to separate tornadoes produced by the same storm system. I am not going to reproduce their revised numbers from memory, and anyone writing this into a document should read that paper directly rather than cite my summary of it.

The word the forecasters were not allowed to use

In 1925 the United States had no tornado warning of any kind, and that was a policy decision as much as a technical limitation. NOAA’s own published histories describe a restriction originating in the 1880s under the Army Signal Service, which then held the national weather service function, barring the word tornado from forecasts on the reasoning that the public would panic and that the forecasts would be wrong often enough to do more harm than good. John Park Finley had been running experimental tornado forecasts for Signal Service districts in that period and had compiled a substantial catalog of tornado events, and the restriction effectively ended that line of operational work. The civilian Weather Bureau that took over forecasting in 1891 carried the policy forward.

The observing system of the day would not have supported much anyway. Forecasting rested on surface observations telegraphed to a central office and plotted on maps a couple of times a day, there was no radar of any kind, and a routine upper-air observing network did not exist until the radiosonde came into regular use in the later 1930s. The ingredients that a forecaster now reads off a sounding, meaning the instability, the shear and the low-level moisture return, were not measurable on a schedule that would have helped anyone on a Wednesday afternoon in March.

Accounts of the morning forecast covering the region describe language about rain and shifting winds, with nothing resembling a severe weather statement, which is what the rules of the time required. The practical consequence is worth stating in plain terms rather than as an abstraction, because the schools, the mines, the mills and the farms along a 200-mile corridor operated that day on a normal schedule with no information at all, and the first notice any of those communities had was the sound and the pressure change a minute or two out.

What actually changed, and when it changed

A tidy version of this story circulates in presentations, and it is a myth: the claim that the Tri-State Tornado produced the modern warning system. The chronology does not support it. The restriction on the word tornado in forecasts remained in place for another thirteen years after 1925, with NOAA histories dating its formal removal to 1938, and public tornado forecasting did not begin for roughly another decade and a half after that. Whatever outrage the 1925 event generated did not convert into a forecasting program in any documented, traceable way, and if you are going to tell the story to a class, tell it with that gap in it.

What did move the needle came out of the military and out of the war. On March 20, 1948 a tornado struck Tinker Air Force Base in Oklahoma, and five days later, on March 25, Air Force officers Ernest Fawbush and Robert Miller issued a forecast of tornado conditions for the base and a tornado occurred, which is generally treated as the first successful operational tornado forecast in the United States. The Weather Bureau stood up a severe local storms unit in 1952 and began issuing public tornado forecasts that year, the function later moved to Kansas City and evolved into what is now the Storm Prediction Center in Norman, Oklahoma, and surplus military radar gave the Bureau its first network detection capability through the 1950s. Doppler capability arrived operationally with the WSR-88D network in the early 1990s, which is what made it possible to see rotation inside a storm rather than infer it from an echo shape.

The 1925 event’s real contribution to doctrine is as the benchmark case. It sets the upper bound in the climatology, it is the event every long-track tornado gets compared against, and it is the reason researchers keep going back to re-survey a path that nobody alive witnessed. Harold Brooks and Charles Doswell have published analyses of United States tornado death rates showing a long decline across the twentieth century that they attribute to warnings, communications and construction rather than to any change in the storms, and the 1925 toll is the starting point that decline is measured from.

The long track as a hazard model

Most tornado planning treats the hazard as a point, which is reasonable because most damaging tornadoes affect one community, one response area and one incident command. A long-track tornado is a line that crosses county boundaries, dispatch center boundaries, state boundaries, mutual aid compact boundaries and National Weather Service forecast office boundaries, and in 1925 it crossed three states in an afternoon. Nothing in that geometry is unusual for a river flood or a hazardous materials release moving downstream, and emergency managers handle those as sequential-impact problems as a matter of routine, yet tornado annexes are usually written as though the damage will be in one place.

The consequence that catches agencies out is the inversion of mutual aid. Your automatic aid partners are, by design, the departments closest to you, and a long-track tornado moving at 60 miles per hour reaches them in minutes. Murphysboro, De Soto, Hurst, Bush and West Frankfort sit along a line, and the resources any one of them would have called were being destroyed in sequence. A tornado that stays on the ground for an hour and a half does not take one jurisdiction out of service, it takes the whole local aid network out of service along a corridor, and the help that is actually available has to come from outside the axis of storm motion rather than from the nearest station.

The flip side of the same geometry is the only good news in it, because forward speed converts distance into time. A community 20 miles ahead of a confirmed tornado moving at 60 miles per hour has roughly 20 minutes, which is more lead time than the national average tornado warning provides. That is the entire function of a warning system stated in one sentence: it turns the geography between the storm and the next town into usable minutes. In 1925 the telegraph could physically have outrun the storm, and no process existed to make anyone send that message or to tell the receiving operator what to do with it.

Infrastructure follows the same corridor the storm does

Rail lines, telegraph and telephone lines, highways and water mains in 1925 followed the same river valleys, ridges and settlement patterns that the towns followed, which means the storm took out the reporting infrastructure in the same stroke that it created the thing that needed reporting. Contemporary accounts describe communities that were unable to send word out for hours, with information travelling by rail crew and by automobile until wires were restored. Modern systems are more diverse than that, though the dependency has not disappeared, since cell sites along a damage path lose commercial power and then lose backup, microwave paths lose line of sight when a tower comes down, and the dispatch center may be in the path along with everything else.

Fire after the tornado is the second compounding failure and it is well documented at Murphysboro, where blazes broke out in the wreckage while the water system was broken and the fire apparatus was trapped or destroyed, and part of the town burned after the wind had passed. Overturned coal and wood stoves were the ignition source in 1925. The modern equivalent is natural gas, because a tornado shears service lines off foundations across a wide area at once, and the utility’s ability to isolate a neighborhood depends on valves and crews that may be behind debris. Any tornado annex that does not name who requests gas isolation, and from whom, is missing a step that will be needed in the first hour.

Medical response along a line has its own arithmetic. The hospitals nearest the damage are frequently in the damage, as happened at Joplin, Missouri on May 22, 2011 when St. John’s Regional Medical Center was struck and had to evacuate while the city was generating casualties. The patients from a 200-mile path do not arrive at one receiving facility, they arrive at a dozen small ones across three states with no single coordinating authority, and search of a narrow corridor hundreds of miles long cannot be run as one incident. In my own years working fire service communications and logistics, the lesson that kept repeating is that the first operational period of a wide-area event is spent finding out what you have lost, and a long track multiplies that by the number of jurisdictions on the line.

Counting a track: why continuity is hard to establish

The question of whether a long damage path was made by one tornado or by several is genuinely hard, and it is not an academic quibble. A supercell can cycle, with one mesocyclone occluding and a new one forming downstream, producing a series of tornadoes whose damage paths overlap or leave only a short gap, and from the ground a day later the result can look like a single continuous swath. The 2013 reassessment of the 1925 path by Johns, Burgess, Doswell and their co-authors is exactly this kind of work applied ninety years late, using period newspapers, damage accounts and site visits to test whether the traditional continuous track holds up along its whole length.

The same problem arose in public view on December 10 and 11, 2021, when the tornado that devastated Mayfield, Kentucky was initially reported as a single track exceeding 200 miles across four states. National Weather Service damage surveys afterward determined that the event was a family of tornadoes rather than one, with the longest continuous track running approximately 165 miles from northwestern Tennessee through western Kentucky. The revision did not make the event smaller and it did not change anybody’s response, and it changed the record substantially, which is why the survey process exists.

For a working agency the practical point is about what you repeat in the first 48 hours. Path length, width, rating and the count of tornadoes in an event are all preliminary until the Weather Service completes its damage survey and publishes the storm data entry, and preliminary numbers get revised in both directions. If your public information officer is quoting a track length on day two, it should be attributed to the forecast office and labeled preliminary, because the correction will circulate far less widely than the original number did.

Warning doctrine now, and where it still breaks

The architecture that replaced silence has a shape worth knowing. The Storm Prediction Center issues convective outlooks days ahead and tornado watches covering broad areas for a period of hours, local forecast offices issue tornado warnings for specific storms, and since October 1, 2007 those warnings have been issued as storm-based polygons rather than for whole counties, which cut the area warned for a given storm substantially. The tornado emergency wording, used for a confirmed large and violent tornado moving into a populated area, was first issued by the Norman forecast office on May 3, 1999 during the Bridge Creek and Moore tornado and is now part of the impact-based warning format. Delivery runs through NOAA Weather Radio, broadcast and cable via the Emergency Alert System, and Wireless Emergency Alerts to mobile phones, which became available in 2012.

The remaining failures are mostly on the human side of the chain rather than the meteorological side. The Weather Service has reported average tornado warning lead times in the general range of about thirteen minutes in recent decades, and verification statistics change, so pull the current figures from NWS performance management rather than quoting a number from a slide deck. What that lead time buys depends on whether people hear it, believe it and have somewhere to go, and the recurring findings after deadly events are that overnight tornadoes kill disproportionately because people are asleep, that manufactured housing residents account for a large share of deaths because the structure offers no protection at any wind speed, and that outdoor warning sirens were designed to alert people who are outdoors and are routinely treated by the public as an indoor alerting system they are not.

Sheltering standards are where the schools of 1925 finally got an answer. The ICC 500 standard, developed by the International Code Council with the National Storm Shelter Association, sets design and construction requirements for storm shelters, and FEMA’s safe room guidance document provides criteria for community and residential safe rooms. Following the National Institute of Standards and Technology technical investigation of the 2011 Joplin tornado, where deaths were reported as 158 and as 161 depending on whether later and indirect deaths are counted, the 2015 edition of the International Building Code added storm shelter requirements for certain new schools and new emergency response facilities in the highest shelter design wind speed region. Adoption varies by state and by local jurisdiction, so verify what your building department has actually adopted rather than assuming the model code applies.

What to do at your agency

• Have the emergency manager confirm with your National Weather Service forecast office which office issues warnings for every part of your jurisdiction, get the warning coordination meteorologist’s name and direct line for your office and for the office covering the counties immediately upwind, and put both entries in the EOC call list this month.
• Add one paragraph to your existing severe weather standard operating procedure naming the role, by title, who notifies the next jurisdiction along the storm track when your agency confirms a tornado on the ground, and specify the method used when telephone and internet are down.
• Have the communications supervisor verify that every NOAA Weather Radio receiver in the comm center and in each station is powered, has fresh backup batteries, and is programmed with the correct SAME county codes, and record the date of the check on the station inspection sheet.
• Pull the mutual aid agreements, mark which partners sit within 30 miles along the prevailing storm motion axis, and begin one new agreement with a department off that axis before the end of the quarter.
• Ask the school district’s safety director for the written severe weather sheltering plan and for a list of which buildings contain a space built to the ICC 500 standard or FEMA safe room criteria, and file the answer with the date it was provided.
• Put a ten-minute scenario on the agenda of a shift briefing that already happens: a confirmed violent tornado 18 miles upwind in the adjacent county, and have the crew state out loud what they do in the next twelve minutes and who they call.
• Have the public information officer confirm that your IPAWS alerting authority, credentials and designated alert originators are current, and complete one test message in the IPAWS test environment with the result logged.

Takeaways

• The Tri-State Tornado of March 18, 1925 killed roughly 695 people according to the National Weather Service and the Grazulis reference catalog, which is the highest toll from a single tornado in United States history.
• The figure of 747 that is often attached to this tornado is a total for all tornadoes across the whole day’s outbreak in several states, and the two numbers disagree because they count different events.
• Murphysboro, Illinois lost roughly 234 people, the largest death toll any single American community has suffered from one tornado, and the school collapse at De Soto killed dozens of children.
• No warning was possible in 1925 partly for technical reasons and partly because of a policy, described in NOAA’s published histories as originating in the 1880s, that barred the word tornado from official forecasts until it was removed in 1938.
• The claim that this disaster produced the modern warning system is a myth, because public tornado forecasting did not begin until the Weather Bureau stood up its severe local storms unit in 1952, following the successful Tinker Air Force Base forecast by Fawbush and Miller in March 1948.
• A long-track tornado is a sequential-impact hazard crossing many jurisdictions, which means your nearest mutual aid partners are likely to be struck in the same event and the help you can actually use has to come from off the storm axis.
• Forward speed converts distance into warning time, so a town 20 miles ahead of a tornado moving at 60 miles per hour has roughly 20 minutes if a process exists to tell them.
• Path length, tornado count and damage ratings are preliminary until the Weather Service survey is complete, as the December 2021 Mayfield event showed when an initially reported single track beyond 200 miles was resolved into a tornado family with a longest continuous segment of approximately 165 miles.