Rail failure statistics from FRA data show that U.S. railroads reported 990 derailments in 2025 (excluding grade crossings), about 1.85 per million train-miles. On main track, broken rails and welds caused 384 of 1,145 track-caused derailments from 2016 to 2025, more than any other track cause group.
This article uses FRA's public datasets as they stood on October 24, 2026. Complete years run from 2015 to 2025. The 2026 figures cover January to July only, so they are a year-to-date reading, not an annual result.
Every count below can be rebuilt from FRA's Form 54 unique train accident dataset and Form 55 operational data. Where a number is our own calculation, we say so and show the method.
What does FRA count as a rail failure?
FRA has no single "rail failure" measure. It records accidents on Form FRA 6180.54 (Rail Equipment Accident/Incident), operating exposure on Form FRA 6180.55, and inspection findings in its enforcement program. A track owner's internal rail test results and service failures are a fourth record, kept by the owner under 49 CFR 213.237 and 213.241.
| Measure | What it represents | Where it comes from |
|---|---|---|
| Derailment | An accident outcome above the year's damage reporting threshold | Form 6180.54 (Form 54) |
| Primary cause code | The cause the reporting railroad assigned, such as T220 | Form 54, FRA cause codes |
| Train-miles | Exposure to railroad activity | Form 6180.55 (Form 55) |
| Inspection defect | A condition found during an FRA or state inspection that doesn't meet a rule | FRA enforcement reports |
| Rail defect found by testing | An internal flaw found by the owner's rail flaw detection | 213.237 test records under 213.241 (not public) |
| Service failure | A broken rail from one of four internal defect types | Owner's 213.237 records (not public) |
Rail failure statistics go wrong when these are mixed. A derailment is a failure that happened. A defect is a condition that someone found, which may or may not have led to a failure.
Key takeaway: count derailments from Form 54, measure exposure from Form 55, and treat inspection and test defects as a separate, earlier signal.
How many train derailments happen in the U.S. each year?
U.S. railroads reported between about 990 and 1,380 derailments a year from 2015 to 2025, not counting highway-rail crossing incidents. The Congressional Research Service published this series in its July 24, 2026 report Freight Rail Safety Issues in the 119th Congress (R47911), using FRA's Rail Safety Overview Report.
| Year | Derailments (CRS) | On main line | On yard, siding or industry track | Unique-accident dataset, Oct 2026 |
|---|---|---|---|---|
| 2015 | 1,351 | 324 | 1,027 | 1,351 |
| 2016 | 1,213 | 272 | 941 | 1,213 |
| 2017 | 1,273 | 335 | 938 | 1,273 |
| 2018 | 1,377 | 320 | 1,057 | 1,376 |
| 2019 | 1,344 | 344 | 1,000 | 1,344 |
| 2020 | 1,117 | 313 | 804 | 1,117 |
| 2021 | 1,118 | 296 | 822 | 1,116 |
| 2022 | 1,229 | 292 | 937 | 1,226 |
| 2023 | 1,310 | 310 | 1,000 | 1,311 |
| 2024 | 1,113 | 295 | 818 | 1,119 |
| 2025 | 976 | 295 | 681 | 990 |
Sources: CRS R47911 Table 3 (July 2026); the last column is our query of FRA's unique train accident dataset on October 24, 2026 (accident type "Derailment").
Two things stand out. Between 70% and 78% of derailments each year happened off the main line, in yards, sidings and industry track. The main-line count has stayed between 272 and 344 a year since 2015, while the yard and industry count drove most of the fall in 2024 and 2025. CRS reached the same conclusion: the drop in the derailment rate in 2024 and 2025 "appears to have been driven by a lower rate of derailments off mainline track."
The two 2025 totals differ (976 against 990) because FRA keeps updating its data as railroads file late or amended reports, and because the two products are compiled differently. Always state the date you pulled the data.
Key takeaway: derailments fell to about 990 in 2025, but main-line derailments, the ones with the highest consequences, have barely moved in a decade.
Why do duplicate Form 54 reports matter?
One physical accident can produce several Form 54 reports. FRA's dataset notes explain why:
"When an accident involves multiple railroads, each railroad must report its data. As a result, there can be multiple records for one accident." FRA, Rail Equipment Accident/Incident Data (Form 54) Subset – Unique Train Accidents, data.transportation.gov
Counting raw Form 54 rows overstates accidents wherever a host railroad, a tenant and a car owner all report the same event. FRA's unique train accident subset keeps one record per accident and also removes highway-rail crossing incidents. CRS adds a matching warning to its own table: main-line and off-main-line counts "may add to more than total derailments" where multiple reports were filed. Use the unique subset for any count, rate or trend.
How do you calculate a derailment rate per million train-miles?
Divide derailments by train-miles and multiply by 1,000,000. A railroad that runs more trains has more chances for an accident, so a rate is fairer than a count for comparing years or railroads.
Derailment rate = derailments ÷ train-miles × 1,000,000
Worked example: national derailment rate, 2021 to 2025
We summed freight, passenger and "other" train-miles from FRA's Form 55 operational data (yard switching miles excluded), then divided the unique-dataset derailment counts by them.
| Year | Derailments | Train-miles (millions) | Derailments per million train-miles |
|---|---|---|---|
| 2021 | 1,116 | 505.5 | 2.21 |
| 2022 | 1,226 | 516.4 | 2.37 |
| 2023 | 1,311 | 531.3 | 2.47 |
| 2024 | 1,119 | 532.9 | 2.10 |
| 2025 | 990 | 534.2 | 1.85 |
For 2025: 990 ÷ 534,200,786 × 1,000,000 = 1.85 derailments per million train-miles. Train-miles rose 5.7% from 2021 to 2025 while derailments fell 11%, so the rate fell more than the count.
This is EMA Industries' calculation, not an FRA-published rate. FRA and CRS use other conventions (CRS reports derailments per 100 million train-miles), and including yard switching miles would lower every rate. Railroads can also edit Form 55 data for up to five years, so totals may shift.
Is 2026 better or worse so far?
The first half of 2026 is running lower than the first half of 2025, but the numbers will rise as late reports arrive. A fair comparison uses the same months in each year:
| January to June | Derailments | Train-miles (millions) | Per million train-miles |
|---|---|---|---|
| 2024 | 545 | 262.2 | 2.08 |
| 2025 | 518 | 266.2 | 1.95 |
| 2026 | 447 | 262.1 | 1.71 |
Source: our query of FRA's unique accident and Form 55 datasets, October 24, 2026. The dataset already holds July 2026, but July is left out here because the most recent month is the least complete. Railroads must file each month's reports "within 30 days after expiration of the month during which the accidents/incidents occurred" (49 CFR 225.11), and amended reports can follow.
Key takeaway: quote rates, not counts, and compare 2026 only against the same months of earlier years.
What is the number one cause of train derailments?
It depends on the track. Across all track from 2016 to 2025, track causes and human factors were nearly tied, at about 36% each. On main track, track causes led (37.5%), and within them broken rails and welds were the largest group.
| Cause group (first letter of FRA cause code) | All track, 2016–2025 | Share | Main track, 2016–2025 | Share |
|---|---|---|---|---|
| Track, roadbed and structures (T) | 4,398 | 36.4% | 1,145 | 37.5% |
| Human factors (H) | 4,319 | 35.7% | 631 | 20.7% |
| Mechanical and electrical (E) | 1,647 | 13.6% | 824 | 27.0% |
| Miscellaneous (M) | 1,536 | 12.7% | 437 | 14.3% |
| Signal and communication (S) | 185 | 1.5% | 14 | 0.5% |
| Total | 12,085 | 3,051 |
Source: our query of FRA's unique train accident dataset (derailments, primary cause), October 24, 2026.
This matches the benchmark study. Liu, Saat and Barkan analyzed FRA derailment data for 2001 to 2010 in Transportation Research Record 2289 (2012) and found that "broken rails or welds were the leading derailment cause on main, yard, and siding tracks". They also found that track and human-factor causes dominated below 10 mph, and equipment causes such as bearing failures, broken wheels and axle defects dominated above 25 mph.
The pattern still holds on main track. In 2016 to 2025, broken rails and welds caused 384 main-line derailments. The next-largest track groups were wide gage (T110 to T113, 137) and buckled track or sun kinks (T109, 123).
More recent research looks at how operations change these rates. C. Tyler Dick and Peter F. Swan compared normalized main-line derailment frequency and severity at the four largest U.S. freight railroads for 2018–2023 against 2012–2017 (Journal of Transportation Engineering, Part A, vol. 152, no. 10, October 2026). They report that rates rose for some causes linked to train make-up, handling and human factors, fell for others, and that cars derailed per incident correlates with train size.
Key takeaway: in yards, people and gage cause most derailments; on the main line, the rail itself is the biggest track problem.
Why are broken-rail derailments more severe?
Broken-rail derailments on main track derailed more cars and did more damage than the average main-line derailment. That is the frequency-versus-severity point made in the risk literature, shown here with FRA data.
| Main-line derailments, 2016–2025 | Number | Cars derailed per derailment | Reported damage per derailment |
|---|---|---|---|
| Broken rails and welds (12 "Broken Rail" codes) | 384 | 11.1 | about $724,000 |
| Human-factor causes (H) | 631 | 6.5 | about $431,000 |
| All causes | 3,051 | 7.3 | about $535,000 |
Source: our calculation from FRA's unique train accident dataset. Cars derailed = loaded and empty freight and passenger cars. Damage is the railroad-reported total in nominal dollars, not adjusted for inflation.
FRA data don't record why, but a likely reason is speed: a main-line rail breaks under trains running at track speed, while many derailments from other causes are slower moves. The worst case is a train of loaded hazmat tank cars. The NTSB's investigation of the October 20, 2006 derailment of 23 ethanol tank cars at New Brighton, Pennsylvania found:
"the probable cause of the derailment … was the Norfolk Southern Railway Company's inadequate rail inspection and maintenance program that resulted in a rail fracture from an undetected internal defect." NTSB, Railroad Accident Report RAR-08/02, adopted May 13, 2008
The same report named FRA's "inadequate oversight of the internal rail inspection process and its insufficient requirements for internal rail inspection" as a contributing factor.
Which FRA T2xx cause codes cover broken rails and welds?
FRA codes rail and joint causes in the T2xx group of its accident cause codes. Twelve of them begin "Broken Rail" in FRA's dataset. Here they are with ten years of derailment counts.
| Code | FRA description (shortened) | Main track, 2016–2025 | All track, 2016–2025 | 213.237 service-failure defect? |
|---|---|---|---|---|
| T207 | Detail fracture from shelling or head check | 110 | 419 | Yes |
| T220 | Transverse or compound fissure | 57 | 174 | Yes |
| T221 | Vertical split head | 55 | 171 | Yes |
| T202 | Base | 53 | 177 | No |
| T204 | Weld (field) | 34 | 50 | No |
| T210 | Head and web separation, outside joint bar limits | 34 | 151 | No |
| T201 | Bolt hole crack or break | 21 | 63 | No |
| T212 | Horizontal split head | 10 | 42 | No |
| T211 | Head and web separation, within joint bar limits | 5 | 24 | No |
| T203 | Weld (plant) | 2 | 5 | No |
| T208 | Engine burn fracture | 2 | 10 | No |
| T218 | Piped rail | 1 | 1 | No |
| Total broken rails and welds | 384 | 1,287 |
Source: our query of FRA's unique train accident dataset (derailments, primary cause), October 24, 2026. FRA marks several of these descriptions as changed over time, so check the cause code guide in force for the year you study.
Three points for track engineers:
- Detail fractures (T207) are the single largest broken-rail cause, on main track and overall. They start at or near the head surface from shelling, head checks or flaking, which is also where surface damage can mask them from ultrasonic beams. See our guide to detail fracture development.
- Welds matter on main track. Field welds (T204) caused 34 main-line derailments, compared with 2 for plant welds. Field thermite welds need their own inspection; see aluminothermic weld inspection.
- Not every T2xx code is a broken rail. T205 and T206 (crossties, spikes and fasteners), T213 to T216 (joint bars and bolts) and T222 (worn rail) sit in the same group but aren't counted above. T101 and T109 are geometry codes, not rail defects.
These are accident cause codes. They aren't the defect names and remedial actions in 49 CFR 213.113, which has no numeric codes. Our guide to FRA defect codes explains the difference.
How does 213.237 rail testing tie to service-failure rates?
Under 49 CFR 213.237, track owners must test rail often enough to keep service failures below a set rate per mile per year. A service failure is a broken rail caused by one of four internal defects, and the rule is built around measured outcomes rather than a test date alone.
Section 213.237(j) defines it this way: "Service failure means a broken rail occurrence, the cause of which is determined to be a" compound fissure, transverse fissure, detail fracture or vertical split head. Those four defects map straight onto three accident cause codes: T207, T220 and T221.
| Track (per rail inspection segment) | Maximum service failures per mile per year |
|---|---|
| Class 4 and 5 | 0.10 |
| Class 3, 4 and 5 with regularly scheduled passenger trains or a hazmat route | 0.09 |
| Class 3, 4 and 5 with regularly scheduled passenger trains and a hazmat route | 0.08 |
Source: 49 CFR 213.237(a) on eCFR. Rates are calculated within 45 days of the end of each 12-month period.
The link between testing and failures works in both directions:
- Testing removes defects before they break. On Class 4 and 5 track, and on Class 3 with scheduled passenger trains or a hazmat route, tests may be no more than 370 days or 30 million gross tons (MGT) apart, whichever comes first.
- The failure rate checks whether the interval is working. If a segment misses its target, the owner must notify FRA within 45 days. After two consecutive years over target, the owner must cut the interval to 10 MGT or drop to Class 2 speeds in the worst area (213.237(d)).
- Unsearched rail doesn't count. Rail where a valid search couldn't be made must be retested, held to 25 mph or replaced before the deadline.
Our full guide to 49 CFR 213.237 covers each paragraph.
Worked example: a segment over its target
A 40-mile Class 4 segment on a hazmat route has a limit of 0.09 failures per mile per year, which allows 3.6 service failures in 12 months. It records 4 (two detail fractures, one transverse fissure and one vertical split head): 4 ÷ 40 = 0.10, over the limit.
The owner must notify FRA within 45 days and may submit a remedial plan. If the next 12 months are also over target, the 10 MGT interval or Class 2 restriction applies. On a 40 MGT line, that is a test about every three months instead of about every nine months.
What the national data says about testing
On main track, the four service-failure defects caused 222 of the 384 broken-rail derailments from 2016 to 2025 (58%). Those are the defects that internal testing intervals are designed to catch. The other 42% came from base breaks, welds, bolt holes and head-web separations. FRA counts none of these as service failures, so a segment can meet its 213.237 target while still having those breaks.
Main-line broken-rail derailments ran between 31 and 44 a year over the decade (38 in 2025), with no clear downward trend. The national data can't show the service-failure rate itself. A broken rail that doesn't derail a train never appears in Form 54, and segment service-failure records stay with the owner.
Key takeaway: service-failure rates are the regulatory test of whether rail testing is frequent enough, and the same four defects account for most broken-rail derailments on the main line.
How do inspection defects relate to derailments?
Inspection defects are leading indicators, and derailments are outcomes; they shouldn't be added together or ranked as one. FRA and state inspectors record defects where a condition doesn't meet a federal rule. FRA's FY2025 enforcement program included more than 70,000 inspections, according to Progressive Railroading's report on FRA's FY2025 Annual Enforcement Report (June 9, 2026).
Raw defect totals mostly reflect how much was inspected. A large railroad with more track, cars and locomotives will always show more defects. Useful comparisons normalize them:
- defects per inspected unit
- defects per inspection day
- track defects per track-mile inspected, by class
- defects compared with derailments of the matching cause group, with a one-year lag
A defect found by an owner's rail flaw test is different again. It goes into the 213.241 record with its size and the 213.113 remedial action, such as a speed restriction, joint bars or replacement. It never appears in FRA's accident data unless the rail later breaks under a train. See our guide to detecting rail defects for how those defects are found.
What can't rail failure statistics from FRA data tell you?
FRA data are strong on what happened and weak on why it happened in that place. Before drawing conclusions, account for these limits:
| Limit | Why it matters |
|---|---|
| One primary cause per accident | Form 54 also has a contributing-cause field, but most summaries, including these tables, use the primary cause only |
| The railroad assigns the cause | The code reflects the reporting railroad's judgment of the cause |
| Damage reporting threshold | Events below the year's monetary threshold aren't in Form 54, so minor yard derailments are undercounted |
| Cause descriptions have changed | Many T2xx descriptions are flagged "Desc. Changed"; long trends can mix definitions |
| No track-mile or tonnage denominator by class | You can't calculate broken rails per mile of Class 4 track from public data |
| Service failures and test results aren't public | Broken rails that don't derail a train, and defects found by testing, sit in owners' records |
| Late and amended reports | Recent months and years rise over time; 2025 grew from 976 (CRS) to 990 |
| Nominal dollars | Damage per derailment isn't inflation-adjusted, so cost trends overstate growth |
| Transit not connected to the general system | Rapid transit outside FRA jurisdiction generally isn't in these datasets; state safety oversight (the CPUC in California) applies |
A rail failure statistic is only as good as the question it answers. For a single segment, the owner's own test, defect and service-failure records will say far more than any national table.
How EMA Industries applies this
EMA Industries performs ultrasonic rail flaw testing and rail weld inspection for transit agencies, short lines and industrial track owners. We're based in Southern California and mobilize nationwide. Owners who track their own service failures and broken rails by defect type, in the same T207/T220/T221 terms FRA uses, can see whether their testing interval is working long before a failure becomes a derailment. To plan a testing program or check field welds on a segment with a rising failure rate, see our ultrasonic rail testing services or contact us.
Related reading: Broken rail causes, defects and prevention · Transverse fissure in rail · Rail testing frequency requirements · What causes a train to derail · Railroad failure analysis
Frequently asked questions
How many train derailments were there in the U.S. in 2025?
FRA's unique train accident dataset showed 990 derailments in 2025 as of October 2026, excluding highway-rail crossing incidents. CRS reported 976 from an earlier FRA overview report. About 295 were on main track.
What is the leading cause of train derailments?
Across all track, track causes and human factors are nearly tied at about 36% each (FRA data, 2016–2025). On main track, track causes lead, and broken rails and welds are the largest track cause group, matching Liu, Saat and Barkan's 2001–2010 finding.
What is the difference between a broken rail and a service failure?
A service failure is a broken rail caused by a transverse fissure, compound fissure, detail fracture or vertical split head (49 CFR 213.237(j)). Broken bases, welds and bolt holes are broken rails but not service failures, and a service failure counts whether or not a train derails.
Where can I download FRA derailment data?
From FRA's safety data pages on data.transportation.gov, linked from safetydata.fra.dot.gov. Use the "Unique Train Accidents" Form 54 subset for counts and Form 55 operational data for train-miles.
Why do derailment numbers differ between sources?
Sources differ on duplicate reports, crossing incidents, the date the data were pulled and the exposure measure. Check whether a figure uses unique accidents, which year's data release it comes from, and whether its rate includes yard switching miles.
Sources
- Rail Equipment Accident/Incident Data (Form 54) Subset – Unique Train Accidents (Not at Grade Crossings), FRA, data.transportation.gov, queried October 24, 2026 (data through July 2026)
- Injury/Illness Summary – Operational Data (Form 55), FRA, queried October 24, 2026
- FRA Safety Data landing page and FRA Office of Safety Analysis
- FRA Rail Safety Overview Report
- Congressional Research Service, Freight Rail Safety Issues in the 119th Congress, R47911, July 24, 2026
- Liu, Saat and Barkan, "Analysis of Causes of Major Train Derailment and Their Effect on Accident Rates," TRR 2289 (2012)
- Dick and Swan, "Temporal Trends in Freight Train Derailment Rates and Severity under Changing Railway Operations," J. Transp. Eng. Part A, 152(10), 2026
- NTSB Railroad Accident Report RAR-08/02, New Brighton, PA (2008)
- 49 CFR 213.237, Inspection of rail, eCFR
- 49 CFR 213.113, Defective rails, eCFR
- 49 CFR 225.11, Reports of accidents/incidents, eCFR
- Track Safety Standards; Improving Rail Integrity, final rule (January 24, 2014), Federal Register
- Progressive Railroading, "FRA collected increased rail safety fines in 2025" (June 9, 2026)