Horizontal Split Head: What It Is and How Ultrasonic Testing Finds It

A horizontal split head is an internal rail defect defined in 49 CFR 213.113(d)(11): a crack inside the rail head, usually 1/4 inch or more below the running surface, that grows horizontally. A 0° ultrasonic probe finds it as an early echo with a lost base echo. One over 4 inches long limits speed to 30 mph.

This guide is for track supervisors, transit engineers and rail flaw test operators who have to recognize the defect, size it and pick the right remedial action. It covers the federal definition, where the defect forms, how it differs from shelling and head and web separation, what the ultrasonic signals look like, a worked timing and sizing example on 136RE rail, and the limits of the method.

What is a horizontal split head under 49 CFR 213.113?

A horizontal split head (HSH) is a progressive crack that lies in a horizontal plane inside the rail head, as defined in 49 CFR 213.113(d)(11). The FRA definition, as published on Cornell's Legal Information Institute copy of the eCFR, reads in full:

"Horizontal split head means a horizontal progressive defect originating inside of the rail head, usually 1/4 inch or more below the running surface and progressing horizontally in all directions, and generally accompanied by a flat spot on the running surface. The defect appears as a crack lengthwise of the rail when it reaches the side of the rail head." — 49 CFR 213.113(d)(11)

Each part of that sentence matters in the field:

• "Progressive" means the crack grows under traffic. It is a fatigue defect, not a one-time break.
• "Originating inside of the rail head" separates it from defects that start at the surface, such as detail fractures and shells.
• "Usually 1/4 inch or more below the running surface" gives a typical depth, not a fixed limit. The word "usually" matters: a shallower horizontal crack is not automatically something else.
• "Progressing horizontally in all directions" describes a crack that spreads along the rail and across the head at roughly the same depth.
• "Generally accompanied by a flat spot" is the one visual clue on top of the rail. As the steel above the crack loses support, the running surface can flatten over the defect.
• "A crack lengthwise of the rail when it reaches the side of the rail head" is the second visual clue, a horizontal line along the side of the head.

The definition sits in a list of 16 defect definitions in paragraph (d). The rule's other parts set who decides what happens next. Under 213.113(a), a person designated under 49 CFR 213.7 decides whether the track may stay in use, and paragraph (c) holds the remedial action table. The current text of the section is on eCFR, 49 CFR 213.113. It was last amended at 85 FR 63388 (October 7, 2020), which changed only the verification sentence in paragraph (b).

The horizontal split head is also tied to another defect by definition. Section 213.113(d)(3) defines a compound fissure as "a progressive fracture originating from a horizontal split head that turns up or down, or in both directions, in the head of the rail." So an HSH that turns into the transverse plane stops being sized by length and becomes a compound fissure, sized by the percentage of head area it weakens. That shift is covered in the remedial action section below, and in more depth in our article on the compound fissure in rail.

Key takeaway: the HSH is the only 213.113 head defect whose definition is horizontal and internal, and that plane is what decides how ultrasonic testing finds it.

How the industry codes the defect

Railroads record defects with short type codes. FRA research report ORD-11/24, Rolling Contact Fatigue: A Comprehensive Review (November 2011), lists the standard railroad classifications, including HSH (horizontal split head) and HSJ (horizontal split head in a joint) (FRA ORD-11/24, Table 2). In accident reporting, FRA cause code T212 covers a broken rail from a horizontal split head. FRA's train accident data on data.transportation.gov list 10 main-track derailments coded T212 from 2016 to 2025 (pulled October 24, 2026), against 55 for T221 vertical split head and 39 for the two head and web separation codes (T210 and T211) (FRA train accident data). These are counts of derailments where HSH was the primary cause, not counts of defects found in testing.

How and where does a horizontal split head form?

A horizontal split head forms inside the rail head, typically from a longitudinal seam, inclusion or other internal discontinuity that grows horizontally under repeated wheel loading. The federal definition fixes the location (inside the head, usually at least 1/4 inch down). Research and patent sources describe the mechanism.

FRA's ORD-11/24 review explains the role of steel cleanliness in internal defects. In its words, "hard inclusions are associated with subsurface RCF defects, and entrapped hydrogen is responsible for some internal defects." The same report says inclusion stringers "are strongly implicated in the formation of vertical split heads," deep-seated shells and transverse defects from those shells (ORD-11/24, section 4.1.2, p. 64). It credits vacuum degassing and batch testing of new rail, both ultrasonic and destructive, with eliminating many of the vertical split heads and split webs seen in the past.

The FRA review names vertical split heads, not horizontal ones, in that sentence, so it supports the general mechanism (internal discontinuities in the steel) rather than an HSH-specific finding. A rail inspection patent gives the HSH case directly: "A horizontal split head usually originates from a longitudinal seam or inclusion," with growth that "usually occurs rapidly along the length of the inclusion and spreads horizontally" (US 6,600,999 B2, granted 2003). A patent is a technical description by its inventors, not an independent study, so treat the "rapidly" as their observation.

Where in the head the defect starts also matters for testing. A 2020 rail inspection patent states that HSH defects "generally are initiated in the gauge and field side of the rail head," and that they "go unreported unless they are large enough to propagate over the web area and interrupt the zero-degree transducer" (US 10,766,510 B1, granted September 8, 2020). That is the inventor's case for a new probe design, so it is a claim, not a measured rate. It does match the geometry: a 0° probe centered over the web looks down a narrow column of the head.

What drives growth after initiation:

No public FRA or TTCI growth rate for horizontal split heads was found (search, October 24, 2026). Unlike detail fractures, for which FRA and NTSB reports give growth estimates in percent of head area per MGT, the HSH is sized by length, and the table's reinspection notes (F and G) are how the rule handles growth: the defect is measured again and moves to a stricter action if it has grown.

Key takeaway: treat the HSH as an internal, steel-quality defect that can start off-center in the head, which is exactly where a single centered 0° beam is weakest.

How is a horizontal split head different from shelling and head and web separation?

A horizontal split head is internal and horizontal, a shell starts near the gauge corner and follows its curve, and a head and web separation runs along the fillet where the head meets the web. The three are easy to confuse because all three run lengthwise and all three can show a horizontal line on the side of the rail. They differ in depth, in plane and in whether 213.113 gives them a remedial action.

Shelling

TTCI's study of shells on the FAST High Tonnage Loop found that they "originate as cracks in the subsurface of the high rail gage corner 1/4 to 3/8 inch from the gage corner surface in 5- and 6-degree curves," and "propagate in a plane parallel to the gage corner surface" until the shell separates from the head (TTCI Technology Digest TD-10-004, March 2010). So a shell follows the rounded gauge corner, mostly on high rails in curves, while an HSH lies flat inside the head.

The regulatory difference is larger. Shelling has no row in the 213.113 remedial action table. The word appears only in the detail fracture definition, (d)(7), which says detail fractures "may arise from shelled spots, head checks, or flaking." A shell on its own carries no prescribed action; it matters because it can hide a detail fracture below it from the ultrasonic beam. A horizontal split head has its own row and its own actions. For more on shells, see rail shelling and spalling.

Head and web separation

Section 213.113(d)(10) defines head and web separation as "a progressive fracture, longitudinally separating the head from the web of the rail at the head fillet area." It is lower in the section than an HSH: at the fillet, where the underside of the head curves into the web, not inside the head. On 136RE rail the head is about 1-15/16 inches deep, so a head and web separation sits near that depth, while a typical HSH sits between 1/4 inch and about 1 inch down.

Head and web separation shares the HSH row in the remedial action table, along with vertical split head, split web, piped rail and longitudinal defective weld. The actions are the same; the classification still matters for the record under 49 CFR 213.241, which requires the type of each defect found, and for any trend analysis of the rail. Head and web separations are common at joints, which is why the industry codes split them into "joint" and "outside joint" (HWJ and HWO in ORD-11/24's list).

Vertical split head

A vertical split head is the HSH's vertical counterpart, a split "through or near the middle of the head" (213.113(d)(16)). Because its plane is parallel to the 0° beam, ultrasonic testing finds it indirectly and with side-looking beams. It is covered on its own page: vertical split head in rail.

Side-by-side comparison

Key takeaway: depth on the 0° trace is the quickest way to tell an HSH from a head and web separation, and the existence of a table row is what separates an HSH from a shell.

How does ultrasonic testing find a horizontal split head?

Ultrasonic testing finds a horizontal split head mainly with the 0° (straight-beam) channel, which shows an early echo from the crack and a drop or loss of the normal echo from the rail base. Side-looking channels cover the parts of the head that the centered 0° beam doesn't reach.

Why the 0° channel is the main tool

A crack's response depends on the angle between the beam and the crack face. A 0° beam travels straight down from the running surface, so it meets a horizontal crack face-on. An open, flat crack face is close to a perfect reflector for steel-to-air, which is why a horizontal split head gives a strong echo when it sits under the beam. A 70° beam, used for transverse defects, meets the same horizontal plane at a steep angle and reflects most of its energy away from the probe. In typical practice the rail beam set is 0° for horizontal defects and loss of back wall, about 37° to 45° for bolt holes and the web, 70° for transverse defects in the head, and side-looking probes. A 1999 university transportation center report on rail flaw inspection data described the same arrangement on test cars: fluid-filled wheels with 0°, 45° and 70° transducers and a side-looker (MPC 99-106, Jeffrey and Peterson, August 1999). No public source sets one standard North American channel list, so treat this as typical practice.

The two 0° signs

On a sound rail, the 0° channel shows one steady echo: the base of the rail, about 7-5/16 inches down on 136RE. A horizontal split head under the beam changes that trace in two ways:

1. An early reflector. An echo appears inside the head gate, well before the base echo. Its time gives the depth of the crack.
2. Loss of back wall (loss of bottom). The crack blocks sound from reaching the base, so the base echo drops or disappears over the length of the defect.

The two signs together are what an operator looks for. A loss of base echo on its own is not proof of a defect. Lost coupling, a dry or contaminated running surface, heavy surface damage or a worn profile that steers the beam can all kill the base echo with no crack present. The operator has to decide which it is. If the early echo is present, there is a reflector. If neither the early echo nor the base echo is present, the rail over that stretch hasn't been tested, and that matters for the rules: the record must show the location of any track not tested under 213.237(g) (213.241(c)(6)).

The NTSB's investigation of the 2006 New Brighton, Pennsylvania derailment shows why. A testing practice that allowed short stretches of loss of bottom to pass was examined, and FRA's track division told the NTSB that such a search "would not constitute a continuous search for defects as defined by the FRA" (NTSB RAR-08/02). That case was about a detail fracture, not an HSH, but the lesson is the same: a loss of bottom is a signal to explain, not one to ignore.

Side-looking channels

A 0° probe centered on the rail looks at a column of steel roughly above the web. A horizontal split head that starts toward the gauge or field side can grow for some distance before it reaches that column, as the 2020 patent cited above argues. Side-looking channels aim the beam across the head from the side, so a horizontal crack near the gauge or field edge shows as an indication on those channels before it interrupts the 0° base echo. A larger HSH can also show on other channels, because a wide horizontal crack can block or redirect the angled beams that would normally cross that part of the head.

Confirming and sizing by hand

Test vehicles and carts flag suspected locations. Under 213.113(b), the owner then has to verify the indication within four hours, or immediately if the suspected defect would need remedial action A, A2 or B. For an HSH, that "immediately" applies when the crack may be over 4 inches long (action B) or broken out in the head (action A). Verification is normally a hand ultrasonic test by a qualified operator, together with a look at the running surface for the flat spot and at the side of the head for a lengthwise crack. Owners using continuous testing under 49 CFR 213.240 are exempt from 213.113(b) and follow that section's verification windows instead.

The hand test does three things:

• Confirms the reflector with a 0° probe at the marked location, checking that the early echo is real and not a surface or geometry signal.
• Measures depth from the echo time, which confirms the defect is inside the head (an HSH) rather than at the fillet (a head and web separation).
• Measures length by moving the probe along the rail and marking where the echo starts and stops, using the edge method the owner's procedure specifies.

Length is what decides the remedial action, so the sizing method matters. The worked example below shows how much.

How well vehicle testing found HSH in benchmark trials

The only public benchmark we found is TTCI's 1998 evaluation of six North American hi-rail test systems on the Rail Defect Test Facility near Pueblo, Colorado. The flaws included two horizontal split heads, 2 × 1 inches and 3 × 2 inches. TTCI reported that "the reliability of finding horizontal split heads was 92 percent" (TTCI TD-99-006, Garcia and Reiff, February 1999). That is 11 hits in 12 runs over the same two flaws, so the 95% interval is wide (roughly 65% to 98%), and repeat runs over one flaw are not independent. Read the figure as "large HSH defects were found almost every time in 1998 trials," not as a detection rate for today's equipment.

Key takeaway: the early echo tells you there is a horizontal reflector and where it is; the loss of bottom tells you how far it extends, but only once coupling has been ruled out.

Worked example: depth, timing and length on 136RE rail

The depth of a horizontal split head comes from the 0° echo time, d = v × t ÷ 2, and its length comes from probe travel between the echo's edges. This example uses nominal 136RE dimensions and a longitudinal velocity in rail steel of 0.2320 in./µs (about 5,890 m/s). Times are steel-only round trips; a wheel or wedge delay is excluded, and real instruments add it.

Inputs

Step 1: expected echoes on sound rail

Step 2: depth of the indication

d = 0.2320 × 5.60 ÷ 2 = 0.650 in.

That is more than 1/4 inch below the running surface and about 1.29 inches above the bottom of the head (1.9375 − 0.650 = 1.2875), so the reflector is inside the head, about a third of the way down (0.650 ÷ 1.9375 = 33.5%). It fits the HSH definition and is too shallow to be a head and web separation, which would echo near 16.7 µs.

Step 3: length and the remedial action

The operator scans along the rail with a hand 0° probe and marks the two points where the early echo falls to half its peak (the 6 dB drop method). The marks are 3.75 inches apart. The 213.113 table puts an HSH of more than 2 and not more than 4 inches under I and G: limit speed to 30 mph (or the class maximum, if lower) and reinspect within 30 days.

Now suppose the procedure instead marks the points where the echo disappears completely. Those points might be 4.40 inches apart, because the beam still catches the crack edge after its center has passed it. That length is over 4 inches and falls under B: a speed set by a designated person with at least one year of supervisory experience, no more than 30 mph, with verification required immediately.

The defect is the same in both cases, but the remedial action differs. The point is not that one method is right. The owner's written procedure must name the sizing method, operators must use it every time, and when a measurement falls close to a 1, 2 or 4 inch boundary, the conservative choice is the more restrictive action until a second measurement settles it.

Step 4: what the record needs

Under 213.241(c), the inspection record must show the date, the track inspected, the location and type of the defect (HSH, not "split head"), its size if not removed before the next train, and the initial remedial action and its date. In this example the entry would read as a 3.75 in. horizontal split head at 0.65 in. depth with remedial action I and G, and a reinspection due within 30 days.

Key takeaway: check every depth against the section's head depth and every length against the table boundaries, and write down which sizing method produced the number.

What remedial action does 213.113 require for a horizontal split head?

For a horizontal split head that is not replaced or repaired, the 213.113(c) table sets the action by length: more than 1 to 2 inches H and F, more than 2 to 4 inches I and G, more than 4 inches B, and a break out in the rail head A. The row is shared with vertical split head, split web, piped rail, head and web separation and longitudinal defective weld. The table is published as an image on eCFR and Cornell, so the values below were read from the Federal Register text of the 2014 rule (79 FR 4256–4257). The 2020 amendment changed only the verification sentence in 213.113(b).

The notes, as the rule states them

• A. "Assign a person designated under § 213.7 to visually supervise each operation over the defective rail." Note A is supervision, not a stop order and not a fixed speed. Each train or move over the rail must be watched by a designated person.
• B. Limit operating speed to that "authorized by a person designated under § 213.7(a), who has at least one year of supervisory experience in railroad track maintenance," and never over 30 mph or the 213.9 maximum for the class, whichever is lower.
• F. Inspect the rail within 90 days. If the rail stays in track, the cycle starts over with each reinspection unless the defect has grown into a more restrictive action. If it isn't reinspected in 90 days, speed is limited to Class 2 or the class maximum, whichever is lower, until it is.
• G. The same as F with a 30-day interval.
• H. Limit speed to 50 mph or the 213.9 maximum for the class, whichever is lower.
• I. Limit speed to 30 mph or the 213.9 maximum for the class, whichever is lower.

The full table and all notes A to I are explained in our guide to FRA defect codes and the 213.113 table.

Three points that are often missed

The speed limits only bite where the track runs faster. For freight on Class 3 track, which is limited to 40 mph under 213.9, note H's 50 mph changes nothing; the "or the class maximum, whichever is lower" wording keeps the lower number. Note I's 30 mph does reduce Class 3 freight speed, and the reinspection notes still apply on every class.

The reinspection clock is part of the action. Under F and G, a missed reinspection is itself a speed restriction. A defect left in track under H and F with no reinspection by day 90 drops the rail to Class 2 speed until someone measures it again.

A turn changes the row. If the reinspection shows the crack has turned up or down, the defect is now a compound fissure under 213.113(d)(3). The compound fissure row is sized by the percentage of the head area weakened: 5% to under 70% is B, 70% to under 100% is A2, and 100% is A. The defect then needs a head-area estimate from the angled channels, not a length.

The rail flaw test operator, qualified under 49 CFR 213.238, finds and sizes the defect; the person designated under 213.7 applies the action. On Class 6 and higher track, the equivalent table is in 49 CFR 213.337.

Key takeaway: an HSH is never just "a split head." Its length band sets the speed and the reinspection date, and a turn moves it to the compound fissure row.

What are the limits of ultrasonic testing for this defect?

Ultrasonic testing is the main way to find a horizontal split head, but it can miss one that is small, off-center or hidden under surface damage, and it can mis-size one near a length boundary. These are the limits a Level III would expect an operator and an owner to plan around.

Off-center starts. A 0° channel over the web doesn't look at the outer parts of the head. An HSH that starts at the gauge or field side may only show on side-looking channels until it grows toward the center. If a test system has weak side-looker coverage, early HSH defects can pass.

Small defects. The rule's table starts at more than 1 inch, but a 1 inch crack under a moving wheel probe gives only a short run of indications. Pulse spacing at speed, gate settings and the operator's call all affect whether a short HSH is flagged. The 1998 benchmark used defects of 2 inches and 3 inches; there is no public data on detection of smaller ones.

Surface conditions above the crack. Shells, head checks, engine burns and the flat spot itself can scatter the beam, weaken coupling or create noise in the head gate. The FRA preamble to the 2014 rule noted the "limitations of current technology to detect internal rail flaws beneath surface conditions." Over badly damaged rail, a loss of bottom may be untested rail rather than a defect, and the operator has to say which.

Telling loss of bottom from lost coupling. This is the most common judgment call. A test car's coupling can drop out for a few inches over a dirty or lubricated patch. Without an early echo, the loss of base alone doesn't prove an HSH, and a pattern of accepting short base-echo dropouts without explanation is how real defects get missed.

Sizing error. The length of the indication depends on the beam width and the edge method, as the worked example shows. Near the 2 and 4 inch boundaries, the measured band and the true band can differ.

Worn rail. Head wear changes the time to each reflector and moves where the beam enters. On a worn 136RE, an echo at a given time sits at a different place in the remaining head than on new rail, so gates and depth checks must use the actual rail height.

The defect can hide what is under it. A large horizontal crack shields the steel below it from the 0° beam and can block angled beams too. A compound fissure turning down from the HSH, or another defect deeper in the head, may not be measurable until the rail is removed.

HSH is not a service-failure defect. The 213.237 service failure rate counts breaks from transverse fissures, compound fissures, detail fractures and vertical split heads. A break from an HSH alone doesn't count toward that rate, but one that turned into a compound fissure does. So an HSH that is left to grow can end up in the rate as a different defect.

For how these limits compare with other NDT methods on rail, see detection of rail defects using NDT methods, and for an inspector's view of the full defect list, see detecting rail defects: a guide for inspectors.

Key takeaway: plan for off-center starts and for the loss-of-bottom judgment; those two account for most missed or misread HSH calls.

How EMA Industries applies this

EMA Industries performs ultrasonic rail flaw testing for transit agencies, short lines and industrial track owners from its base in Southern California, and mobilizes nationwide. On a horizontal split head, the useful deliverable is more than a mark on the rail: the depth from the 0° echo, the length by a stated sizing method, the 213.113 remedial action band it falls in, and any stretch where loss of bottom couldn't be explained and must be recorded as not tested. If you are planning a test program or need suspected head defects verified and sized, contact EMA Industries.

FAQ

Is a horizontal split head the same as a shell?

It isn't. A shell starts near the gauge corner, about 1/4 to 3/8 inch below its surface, and follows the corner's curve until a piece breaks out; it has no row in the 213.113 remedial action table. A horizontal split head starts inside the head, usually at least 1/4 inch below the running surface, grows flat in all directions, and has its own length-based remedial actions.

What does a horizontal split head look like on the rail?

The rule names two signs. A flat spot often forms on the running surface over the defect, and when the crack reaches the side of the head it shows as a lengthwise crack along the rail. Many HSH defects show neither sign while they are small, which is why internal testing under 213.237 is needed.

Why does the 0° probe lose the back-wall echo over a horizontal split head?

The crack lies across the beam's path, so most of the sound reflects from the crack face and never reaches the rail base. The trace shows an early echo at the crack depth and a weaker or missing base echo. Lost coupling also removes the base echo, so the early echo is what confirms a defect.

What happens when a horizontal split head turns up or down?

It becomes a compound fissure under 213.113(d)(3). The remedial action is then set by the percentage of head area weakened (B, A2 or A) rather than by length. The angled channels, especially 70°, are used to estimate that area.

Does remedial action A mean the train has to stop?

Action A, which applies to an HSH that has broken out in the rail head, requires a person designated under 213.7 to visually supervise each operation over the defective rail. It is a supervision requirement with no set speed, not a stop order. The designated person decides how each move proceeds.

Sources

• 49 CFR 213.113, Defective rails (Cornell LII copy, read October 24, 2026)
• 49 CFR 213.113 on eCFR
• 49 CFR 213.237, 213.238, 213.240 and 213.241 on eCFR
• FRA final rule, Track Safety Standards; Improving Rail Integrity, 79 FR 4234, January 24, 2014 (remedial action table and notes, pp. 4256–4257)
• FRA final rule, 85 FR 63362, October 7, 2020 (amendment to 213.113(b))
• FRA ORD-11/24, Rolling Contact Fatigue: A Comprehensive Review, November 2011
• TTCI Technology Digest TD-99-006, Benchmarking Results of North American Rail-Flaw Detection Technologies, February 1999
• TTCI Technology Digest TD-10-004, Shell Defect Analysis Using Scanning Electron Microscopy Techniques, March 2010
• Jeffrey and Peterson, MPC 99-106, Assessment of Rail Flaw Inspection Data, August 1999
• NTSB RAR-08/02, New Brighton, Pennsylvania derailment report
• US Patent 6,600,999 B2 (2003), hi-rail rail inspection system
• US Patent 10,766,510 B1 (2020), detecting defects in the head area of rail
• FRA train accident data, data.transportation.gov dataset byy5-w977 (pulled October 24, 2026)