Railway Level Crossing Deaths: What the Evidence Shows
Key takeaways · 11 min read
- US crossing collisions fell from 9,461 in 1981 to 2,051 in 2010, then levelled off; preliminary 2025 figures show 2,274 collisions and 285 deaths.
- Gates and flashing lights explain about a fifth of the 1975 to 2001 decline. Safer roads in general explain about two-fifths.
- In Great Britain, pedestrians at passive crossings made 1% of crossings and suffered 48% of deaths.
- Railway-controlled crossings are the safest per crossing, but they close for longer, and on British economics the delay can outweigh the risk.
In 1981, trains in the United States hit vehicles and people at level crossings 9,461 times, and 728 people died. By 2010 the count had fallen to 2,051 collisions and 260 deaths. That is one of the great quiet safety improvements of the late twentieth century, achieved while both road and rail traffic grew.
Then it stopped. Preliminary federal figures for 2025 show 2,274 collisions and 285 deaths, slightly more than in 2010. Fifteen years of flat numbers, after thirty years of steep decline.
The obvious explanation for the decline is the barrier arm and the flashing lights, and they did help. But the best attempt to take the improvement apart found that gates and lights explain about a fifth of it. This is what the research says about where level crossing deaths come from, which kinds of crossing are safest, and why the safest kind is not always the one a railway chooses.
US highway-rail grade crossing collisions
Collisions per year, all public and private crossings.
Federal Railroad Administration data as published by Operation Lifesaver. Deaths: 728 in 1981, 260 in 2010, 194 in 2020 and 285 in 2025 (preliminary).
Where the decline came from
Shannon Mok and Ian Savage at Northwestern University pooled data from 49 states between 1975 and 2001 and asked how much of the fall in crossing collisions and deaths each factor could account for. Their answer, published in Risk Analysis in 2005, is not what a crossing engineer might hope.
About two-fifths of the decrease came from things that made every road safer, not crossings in particular: less drunk driving and better emergency medical care. The installation of gates and flashing lights accounted for about a fifth. Operation Lifesaver, the public education campaign that began in the 1970s, accounted for about a seventh, and so did the extra “ditch lights” fitted to the front of locomotives in the mid-1990s. Closing crossings, mostly through line abandonments and consolidation of little-used ones, accounted for about a tenth.
A later analysis by Savage found that more Operation Lifesaver activity in a state was associated with fewer collisions between 1996 and 2002. Education is rarely credited with measurable safety effects. Here it has one, at least in these regressions.
What drove the fall, 1975 to 2001
Approximate share of the reduction in US crossing collisions and deaths.
Mok and Savage, Risk Analysis, 2005. Negative binomial regressions on data from 49 states. All shares are approximate: the authors give them as rough fractions.
Why before-and-after studies flatter the equipment
Many older estimates of what a warning device achieves compare collisions at a crossing before and after it was upgraded. Ezra Hauer and Bhagwant Persaud showed in 1987 that this method is biased. Crossings are usually upgraded because they have just had a bad run of collisions. Some of that bad run was chance, and the next few years would have been quieter even with no upgrade at all. Credit that belongs to regression to the mean gets assigned to the gate.
They proposed combining the crossing’s own accident history with what its traffic and design predict, and wrote plainly that the current estimates of warning-device effects were incorrect. Their method became standard in road safety evaluation. It does not mean gates fail. It means a before-and-after figure from a crossing chosen because of its record should be read as an upper bound.
With that caution, the direction of the evidence is consistent. A 1985 Federal Highway Administration study of US crossings from 1975 to 1980 found that ordinary stop signs at passive crossings reduced collisions by an average of 35%. A study of East Japan Railway crossings found 0.59 accidents per million trains at crossings with barriers against 1.25 at crossings with warning bells only, and 0.12 at crossings with obstacle detectors against 0.43 without them.
East Japan Railway: accidents per million trains
By type of crossing protection.
Anandarao and Martland, 1998. Crossings without obstacle detectors: 0.43. Crossings with visibility under 20 metres had a 50% higher mean accident rate.
Who dies, and at which crossings
The most useful recent analysis comes from Great Britain, where the rail infrastructure manager publishes how many vehicles and pedestrians cross each level crossing every day. Andrew Evans and Peter Hughes used those counts to calculate deaths per crossing, not just deaths per year.
Between 2003 and 2017 there were 131 accidental road or footpath user deaths at British level crossings; suicides were excluded. Of those, 97 were pedestrians or cyclists and 34 were occupants of road vehicles. Seventy-three died at passive crossings, which have no lights or barriers, 46 at automatic crossings, and 12 at crossings controlled by railway staff.
The distribution of deaths looks nothing like the distribution of use. Pedestrians at passive crossings accounted for 48% of the deaths and about 1% of the crossings made. Vehicle occupants at railway-controlled crossings made 63% of all crossings and suffered one death in fifteen years.
Per billion crossings, the risk ranged from 0.05 for people in vehicles at railway-controlled crossings to 181 for pedestrians at passive footpath crossings, roughly three orders of magnitude. Automatic crossings carried more than ten times the fatality rate of railway-controlled ones.
None of this makes level crossings the largest danger on a railway. Operation Lifesaver notes that crossing collisions and trespass on the tracks together account for more than 95% of US railroad deaths, and trespass is a separate problem with different causes. The British figures above deliberately leave trespass and suicide out, so they describe only people using a crossing as a crossing.
Great Britain, 2003 to 2017
Accidental road and footpath user deaths at level crossings, by type of crossing. Suicides excluded.
Evans and Hughes, Accident Analysis and Prevention, 2019 (PMID 31128442). Of the 131 deaths, 97 were pedestrians or cyclists and 34 vehicle occupants.
The trade nobody puts on the sign
Railway-controlled crossings are the safest kind. So why does a railway not convert every automatic crossing to railway control?
Evans and Hughes put the answer in pounds. When Britain began replacing staffed crossings with automatic ones from the 1960s, part of the hope was that removing the operator would remove human error. That hope, they write, went unfulfilled, because most of the error at level crossings belongs to road users, not railway staff. Automatic crossings turned out to be riskier per crossing, not safer.
What automatic crossings do is close for less time. The authors estimate a typical automatic half barrier closes for about 50 seconds per train and a railway-controlled crossing for about 150, because signals have to clear well ahead of a train. The delay to road users grows with the square of the closure time, so tripling it multiplies the delay by nine.
For a typical railway-controlled crossing, converting to automatic operation would raise the valued cost of expected deaths by about £23,000 a year and cut the valued cost of delay by about £66,000. On the economics used by the British government, the less safe crossing wins. The authors are careful that this applies to simple, representative crossings and depends on how time and life are valued. But it explains why a known safer design does not spread by itself.
Converting a railway-controlled crossing to automatic
Annual change in value for a representative British crossing.
Evans and Hughes, Accident Analysis and Prevention, 2019. Closure times assumed: 150 seconds railway-controlled, 50 seconds automatic.
The crossing at Mendon
On 27 June 2022 an Amtrak train hit a dump truck at a passive crossing near Mendon, Missouri. The truck driver and three passengers died, and 146 passengers and crew were injured. The crossing had a stop sign and crossbucks and nothing else.
The National Transportation Safety Board found that the driver had not stopped. It also found that the crossing itself made stopping and starting again hard for a heavy vehicle: the road approach was about thirteen times steeper than recommended, and the road met the track at an angle about 30 degrees sharper than the AASHTO design guidance. The crossing has since been closed, and state officials made plans to close more passive crossings and route traffic to active ones.
The NTSB chair, Jennifer Homendy, summed up the board’s position in one line: “The safest rail grade crossing is no rail grade crossing.”
Australia had its own version on 5 June 2007, when a truck hit a passenger train at a crossing near Kerang, Victoria, and 11 people died. A later paper noted that more than half of Victoria’s roughly 2,000 level crossings were controlled only by stop or give-way signs, and that fitting boom barriers and flashing lights to all of them would cost more than A$300 million.
What people do at the barrier
Almost every investigation of a crossing collision points to a road user, and the details matter. In Finland, Sirkku Laapotti studied all 142 fatal motor vehicle accidents at level crossings from 1991 to 2011, each already investigated by accident teams. At passive crossings the typical error was not seeing the train in time, and the surroundings often made it hard to see. At active crossings the typical error was deliberate risk taking.
Robert Raub’s analysis of ten years of US federal data found male drivers outnumbering female drivers three to one, drivers under 25 over-represented, and drivers over 65 more likely than others to be killed. Collisions were more likely in daylight and clear weather, which says something about attention rather than visibility.
Some commonly suggested fixes have disappointing records. A Dutch before-and-after study for the rail infrastructure manager found that rumble strips on the approach to crossings did not reduce speeds, and that extending barriers across footpaths improved pedestrian behaviour but did not stop people who crossed on red deliberately. Evans and Hughes cite work linking longer warning times before a train arrives with higher collision risk: one plausible reading is that lights which flash for a long time with no train in sight teach people to discount them.
What the evidence does not show
It does not show why the US decline stalled after 2010. The flat numbers since then have not, as far as we can find, been taken apart the way Mok and Savage took apart the earlier fall. Traffic growth, crossing inventories and reporting changes could all play a part, and we would be guessing to choose one.
It does not give clean effect sizes for individual devices. Many of the classic estimates come from before-and-after studies of crossings upgraded after bad years, which Hauer and Persaud showed to be biased upwards.
It is not directly comparable across countries. Researchers note that some countries’ crossing data include suicides and trespass, which have different causes, while the British analysis excludes them. The Japanese and Finnish figures cover particular railways and periods.
And the British cost comparison is only as good as the values it uses for time and for life. Change those and the balance between safety and delay moves.
Questions people ask
How many people die at level crossings in the US each year?
Federal figures show 260 deaths in 2024 and 285 in 2025 (preliminary), from about 2,270 collisions a year. The number fell steeply from 728 in 1981 to 260 in 2010 and has been roughly flat since.
Are crossings with barriers safer than crossings with only signs?
Yes. British data show far higher death rates per crossing at passive crossings than at automatic or railway-controlled ones, and Japanese data show fewer accidents per million trains where barriers are fitted. Exact effect sizes vary by method.
Who is most at risk at a level crossing?
In Great Britain, pedestrians at passive footpath crossings, by a wide margin. In US data, young male drivers are over-represented in collisions and older drivers are more likely to die in them.
Why are automatic crossings used if staffed ones are safer?
Because they close to road traffic for much less time. A British analysis found the delay saved by automatic crossings can outweigh, in economic terms, the extra fatality risk.
Do public education campaigns work?
A US analysis attributed about a seventh of the fall in crossing collisions between 1975 and 2001 to Operation Lifesaver, and found more campaign activity associated with fewer collisions. That is regression evidence, not a trial.
The short version
- US crossing collisions fell from 9,461 in 1981 to 2,051 in 2010, then levelled off; preliminary 2025 figures show 2,274 collisions and 285 deaths.
- Gates and flashing lights explain about a fifth of the 1975 to 2001 decline. Safer roads in general explain about two-fifths.
- In Great Britain, pedestrians at passive crossings made 1% of crossings and suffered 48% of deaths.
- Railway-controlled crossings are the safest per crossing, but they close for longer, and on British economics the delay can outweigh the risk.
- Before-and-after studies of upgraded crossings tend to overstate what the upgrade achieved.
This article summarises published research and official statistics on level crossing safety. It is not safety guidance for any particular crossing. At any level crossing, obey the lights, barriers and signs, and never go round a lowered barrier. If you are stuck on a crossing, get out and away from the track, and call the emergency number shown at the crossing if there is one.
Further reading. Evans and Hughes, ‘Traverses, delays and fatalities at railway level crossings in Great Britain’, Accident Analysis and Prevention, 2019, is the study that counts deaths per crossing and prices the trade-off. Mok and Savage, ‘Why has safety improved at rail-highway grade crossings?’, Risk Analysis, 2005, takes the US decline apart. The NTSB report on the Mendon collision (RIR-23-09) is free to read.
- Doom, Niall Ferguson (2021). A historian on why institutions that know what the danger is so often fail to act on it. Broad and opinionated rather than technical.
- The Data Detective, Tim Harford (2020). On asking what a number is out of, which is the whole difference between deaths per year and deaths per crossing.
- Noise, Daniel Kahneman, Olivier Sibony and Cass Sunstein (2021). On the variability of human judgment, and when rules and machines beat people. The crossing data complicate its case, which is a reason to read them together.
Sources
- Federal Railroad Administration. Highway-rail grade crossing collisions, fatalities and injuries by year, as published by Operation Lifesaver (accessed September 2026).
- Mok SC, Savage I. Why has safety improved at rail-highway grade crossings? Risk Analysis, 2005. doi:10.1111/j.1539-6924.2005.00642.x.
- Savage I. Determining the causes of the improvement in grade crossing safety in the United States since 1975. Working paper, Northwestern University.
- Hauer E, Persaud BN. How to estimate the safety of rail-highway grade crossings and the safety effects of warning devices. Transportation Research Board, 1987.
- Farr EH, Hitz JS. Effectiveness of motorist warning devices at rail-highway crossings. Federal Highway Administration, 1985.
- Anandarao S, Martland CD. Level crossing safety on East Japan Railway Company: application of probabilistic risk assessment techniques. 1998. doi:10.1023/a:1005044212685.
- Evans AW, Hughes P. Traverses, delays and fatalities at railway level crossings in Great Britain. Accident Analysis and Prevention, 2019. PMID 31128442.
- Laapotti S. Comparison of fatal motor vehicle accidents at passive and active railway level crossings in Finland. IATSS Research, 2016.
- Raub RA. Examination of highway-rail grade crossing collisions nationally from 1998 to 2007. Transportation Research Record, 2009. doi:10.3141/2122-08.
- van der Horst ARA, Bakker P. The effectiveness of safety measures at railway level crossings on road user behaviour. TNO Human Factors, 2002.
- Hall K, Arvanitakis A, Beer K, Kenos A. Railway level crossing infrastructure countermeasures. 2011 (TRID record 1285899).
- Stefanova T, Burkhardt JM, Filtness A, and colleagues. Systems-based approach to investigate unsafe pedestrian behaviour at level crossings. 2015.
- National Transportation Safety Board. Collision of Amtrak train with dump truck, Mendon, Missouri, 27 June 2022. Railroad Investigation Report RIR-23-09; press release, 2 August 2023.
