A second bond does not put 30 to 50 percent of neutral current on the conduit. A 3 AWG neutral and 8 AWG equipment grounding conductor split 30 A into 7.1848 A on the metal, set by wire area. The worse mistake is the repair: three missing words in 250.6(B)(1) take the fault loop from 310.8808 A to 4.8 A.
If you came from one of our other grounding pages, this is the one about what happens after the jumper is landed. The X0 page already settles which table sizes the jumper, where the single point sits, and what the eight subparts of 250.30(A) each cover. The resistance grounding page already handles the 1,200 A pickup and the charging current. Neither one follows the neutral current after somebody adds a second bond. Here we take two published wire tables, one published 100 ft example and the NFPA public comment reproduction of 250.6, then recompute the split, the heat and the fault loop. No sizing chart, no SDS lecture, no product catalog.
The short answer, before the mistake list
Here is the six row shape search engines like to pull, with every adjective replaced by a quantity you can measure on site before you sign anything.
| If your question is | What actually answers it | Number to check | What goes in the spec |
|---|---|---|---|
| How much current goes on the metal | The resistance ratio of the two parallel paths | 30 x 0.024500 / 0.102300 gives 7.1848 A | Conductor size and length for both paths |
| Is it really 30 to 50 percent | The circular mil areas, not a hazard list | 16,510 / 69,130 gives 23.8825 percent | The installed EGC size, not a percentage claim |
| Where does the rest of the current go | Kirchhoff at the second bond | 22.8152 A stays on the neutral | Clamp readings at both ends of the feeder |
| Can I just lift the ground wire | 250.6(B)(1) includes but not all | 310.8808 A becomes 4.8 A | One redundant connection removed, one kept |
| Does EMI count as the reason | 250.6(D) excludes it | Nine clauses, four permitted moves | A separate power quality study |
| Which section number do I cite | The edition the AHJ adopted | 11 citations, three different numbers | Edition, adoption date, written ruling |
Read the second column top to bottom. Not one row says “dangerous” or “always” or “never.” Every answer is a resistance ratio, an area ratio, an amp reading, a fault current, or a citation count. Two columns of warnings can be correct in every cell and still leave the crew with no way to tell which wire to move on Tuesday morning.
What the ten ranking pages get right
Give the search results their due. One reference site publishes a complete worked example: a 100 ft feeder with 3 AWG copper conductors, an 8 AWG copper equipment grounding conductor, 30 A of unbalanced neutral current, and 7.2 A on the grounding path after a downstream bond. The same page gives 120 V, 0.386 ohm and 25 ohm, then reports 311 A through the metal fault loop against 4.8 A through a ground rod. Those are the only two hard numbers in the whole result set, and both survive a calculator. Our 7.1848 A sits 0.2122 percent below its 7.2 A, and our 310.8808 A sits 0.0383 percent below its 311 A.
A trade magazine supplies the IEEE 1159-2019 definition of a ground loop, quotes 250.6(C), and walks through several separately derived system scenarios. A code commentary walks through all four jumper names and notes that three of them size from a conductor while the fourth sizes from an overcurrent device. An exam page writes out the current divider, I1 over I2 equals R2 over R1, which is exactly right. A specialist site reads 250.6(A) through (E) in order and warns that lifting the equipment grounding conductor is never the repair. So the rules are on the table. What is missing is the arithmetic under them, and in one case the arithmetic is explicitly declined: that trade magazine says its intent is to identify potential loop problems rather than to analyze the exact flow or quantity of current in detail.
Four numbers that do not survive a calculator
| Claim as printed | Where it comes from | What it actually measures | What breaks it |
|---|---|---|---|
| 30 to 50 percent of neutral current on metal | Exam prep diagrams and hazard lists | A range with no conductor sizes behind it | 23.9492 percent on a 3 AWG and 8 AWG pair |
| Disconnect one or more grounding connections | A paraphrase of 250.6(B)(1) | Drops the qualifier but not all | Fault loop goes from 310.8808 A to 4.8 A |
| EMI is the objectionable current | Almost every hazard list | Something 250.6(D) excludes by name | Nine clauses read together |
| 250.24(A)(5) is the load side rule | Older editions and exam pages | One numbering out of three in circulation | 2026 text reported as 250.24(B) |
The first row is the one that sends people looking in the wrong place. A percentage with no wire sizes attached cannot be checked, so it gets repeated. The second row is the one that can hurt somebody, and we come back to it with the fault loop in front of us.
Seven steps, in the order a site walk actually makes them
- Find every neutral to case connection between the source and the last panel, including the green screw in a subpanel and the strap somebody left in a generator.
- Write down the conductor size and the length of each path back to that bond, including metal conduit, building steel and piping.
- Put a clamp meter on the equipment grounding conductor at normal load and record the number.
- Clamp the phase and the neutral together; the net reading should be zero.
- Decide whether the reading is load current on a parallel path or fault current doing its job.
- Remove one redundant connection, never the last one, then read the meter again.
- Record the adopted NEC edition and the section number the AHJ will enforce.
Steps 5 and 6 are where the ranking pages split. One of them treats any current on a grounding path as a reason to open that path. The code says the opposite, and the difference is measurable.
The current is a divider, not a percentage
Two conductors joining the same two points are in parallel. That is the whole mechanism. Once a second neutral to case bond exists, the load current returning from the downstream panel has two roads back to the source: the insulated neutral, and every metal path bonded to the equipment grounding conductor. Neither road knows it was supposed to be a safety path. They split the current in inverse proportion to their resistance.
Two independent ways to get 7.1848 A
Take the published example. Chapter 9 Table 8 gives 3 AWG stranded copper as 52,620 circular mils and 0.245 ohm per 1,000 ft at 75 C, and 8 AWG stranded copper as 16,510 circular mils and 0.778 ohm per 1,000 ft. At 100 ft those become 0.024500 ohm and 0.077800 ohm. The area ratio is 3.1872 and the resistance ratio is 3.1755.
| Method | Input | Arithmetic | Current on the metal | Share of 30 A |
|---|---|---|---|---|
| Resistance divider | 0.024500 and 0.077800 ohm | 30 x 0.024500 / 0.102300 | 7.1848 A | 23.9492 percent |
| Circular mil area | 52,620 and 16,510 cmil | 30 x 16,510 / 69,130 | 7.1648 A | 23.8825 percent |
| Difference between them | Stranded 8 AWG runs slightly high | 7.1848 minus 7.1648 | 0.0200 A | 0.2782 percent of the first |
| Published rounded figure | Given as 7.2 A | 7.2 minus 7.1848 | 0.0152 A | 0.2122 percent above ours |
The neutral keeps 22.8152 A. Carrying the unrounded currents, both paths drop 0.558974 V, because that is what being in parallel means. If you only remember one line from this article, make it the second row. For the same metal and the same length, resistance goes as one over area, so the split is just the smaller area over the sum of the two areas. You do not need a resistance table to predict it.
Why length and temperature drop out
Double the run to 200 ft and both resistances double, to 0.049000 and 0.155600 ohm. The divider returns 7.1848 A again, with a difference of exactly zero. Warm both conductors from 75 C to 90 C and Table 8’s own correction, one plus 0.00323 times 15, scales both by 1.0485. The split is unchanged. Length, temperature and even the copper resistivity constant cancel, because they multiply both resistances by the same factor.
| Variable changed | Neutral resistance | Metal path resistance | Split of 30 A | Change |
|---|---|---|---|---|
| 100 ft at 75 C | 0.024500 ohm | 0.077800 ohm | 7.1848 A | Baseline |
| 200 ft at 75 C | 0.049000 ohm | 0.155600 ohm | 7.1848 A | Zero |
| 100 ft at 90 C | 0.025688 ohm | 0.081573 ohm | 7.1848 A | Zero |
| 3 AWG against 8 AWG | 52,620 cmil | 16,510 cmil | 7.1848 A | Baseline |
The 90 C row works the same way: 0.024500 x 1.0485 gives 0.025688 ohm and 0.077800 x 1.0485 gives 0.081573 ohm, so the ratio stays 3.1755. The only thing that moves the share is the wire size. A buyer who asks how long the feeder is has asked a fair question about voltage drop and an irrelevant one about circulating current. Ask for both conductor sizes instead.
What the meter would say
| Clamp position | Reading on the 100 ft example | What it tells you |
|---|---|---|
| Phase and neutral together | 0 A | Kirchhoff holds at that point |
| Neutral conductor alone | 22.8152 A | Less than the 30 A the load returns |
| Equipment grounding conductor alone | 7.1848 A | Load current on a path meant for faults |
| Phase conductor alone | 30 A | The load current itself |
| Metal conduit downstream of the second bond | A share of 7.1848 A | How many parallel metal paths exist |
That first row is the diagnostic almost nobody prints. If a clamp around phase and neutral reads zero and a clamp around the grounding conductor reads 7.1848 A, the current is not leaking to earth. It found a second way home. The last row is why the number grows in real buildings, which is the next section.
Four names for one current
| Name used on the ranking pages | What it strictly describes | Is it the 7.1848 A |
|---|---|---|
| Objectionable current | NEC 250.6(A), current on paths not meant to carry it | Yes, once the AHJ classifies it |
| Circulating neutral current | A loose label used in exam diagrams | Yes, though nothing circulates in a closed loop |
| Neutral current diversion, or NCD | The meter man’s term for stray neutral current on metal | Yes, and it is the measurable one |
| Ground loop | IEEE 1159-2019, two nominally grounded points joined by a conductor while at different potentials | It is the topology, not the current |
Strictly speaking, no current circulates here. Load current leaves the source on a phase conductor and comes back split between two parallel conductors. A ground loop is the loop of metal that makes the second path possible. The distinction matters because 250.6(D) says currents that cause electromagnetic interference or data errors are not the objectionable currents this section addresses. Almost every ranking page leads with EMI. The code deliberately does not.
What 30 to 50 percent would actually require
If a page says 30 to 50 percent without naming a single conductor, you can reverse the arithmetic and find out what it assumed. Given a share f, the metal path needs an area ratio of f over one minus f against the neutral. On the 3 AWG and 8 AWG example, a real 8 AWG is 31.3759 percent of the neutral area, so the real share is 23.8825 percent by area.
Reverse the arithmetic
| Claimed share | Area ratio needed | EGC area needed | Versus a real 8 AWG | Overstates 23.8825 percent by |
|---|---|---|---|---|
| 30 percent | 0.4286 | 22,551.4286 cmil | 1.3659 times | 1.2561 times |
| 40 percent | 0.6667 | 35,080 cmil | 2.1248 times | 1.6749 times |
| 50 percent | 1.0000 | 52,620 cmil | 3.1872 times | 2.0936 times |
| 60 percent | 1.5000 | 78,930 cmil | 4.7807 times | 2.5123 times |
The 50 percent row is the tell. For half the neutral current to run on the grounding path, that path needs the same copper area as the neutral itself. Table 250.122 does not ask for that. A 100 A overcurrent device gets an 8 AWG copper EGC against 3 AWG phase conductors, which is 31.3759 percent of the area, not 100 percent. So the printed range is not describing this feeder at all.
The 2.3333 times area span
From 30 percent to 50 percent the required area goes from 22,551.4286 to 52,620 cmil, a 2.3333 times span. A range that wide, printed without a single wire size, is not a calculation. It is a way of saying “some.” There is a second reading, though, and it is more useful than calling the range wrong. In a real building the grounding path is not one 8 AWG conductor. It is the EGC plus the metal raceway plus the panel cabinets plus bonded water piping plus structural steel. Add those in parallel and the effective area grows. The 30 to 50 percent range becomes achievable, just not for the reason anyone printed.
What to ask the supplier instead
Do not ask whether a second bond causes 30 percent or 50 percent stray current. Ask for the neutral conductor size, the equipment grounding conductor size on the offered feeder, and a list of every other bonded metallic path running in parallel with it. Those three answers give you a divider you can compute. Then ask for a clamp reading at normal load on the finished job. If the number does not match the divider, there is a path nobody drew.
Seven metal paths, not one green screw
The 100 ft example has exactly one extra conductor in parallel with the neutral. That is a diagram, not a building. Between a real service disconnect and a real subpanel, the current can also ride the metal raceway, the panel cabinets at both ends, an equipment grounding conductor, bonded metal water piping, structural steel, cable tray and a second feeder EGC. Each one is another resistor in parallel, and parallel resistors do not split the current politely. They add conductance.
Add paths in parallel and the share climbs
For this table we use the area method, because the equivalent area of several identical paths is just the sum. One path gives 7.1648 A against the resistance method’s 7.1848 A, a 0.2782 percent difference. The ranking is what matters here, not the fourth decimal.
| Parallel 8 AWG equivalent paths | Share on the metal | Amps of the 30 A | Amps left on the neutral | Change from one path |
|---|---|---|---|---|
| 1 | 23.8825 percent | 7.1648 A | 22.8352 A | Baseline |
| 2 | 38.5567 percent | 11.5670 A | 18.4330 A | 1.6144 times |
| 3 | 48.4875 percent | 14.5463 A | 15.4537 A | 2.0302 times |
| 4 | 55.6548 percent | 16.6964 A | 13.3036 A | 2.3304 times |
| 6 | 65.3085 percent | 19.5926 A | 10.4074 A | 2.7346 times |
| 8 | 71.5106 percent | 21.4532 A | 8.5468 A | 2.9943 times |
| 12 | 79.0141 percent | 23.7042 A | 6.2958 A | 3.3084 times |
Read the last row. Twelve bonded metallic paths take 79.0141 percent of the neutral current, leaving only 6.2958 A on the conductor that was installed to carry it. Eight paths are already 2.9943 times the single path current. Nothing failed, no breaker moved, and the neutral conductor is doing less than a quarter of its job.
What the published range actually needs
Reverse the same arithmetic. To reach a claimed share f with N equal paths, N equals f over one minus f, times the neutral to EGC area ratio of 3.1872.
| Claimed share | Paths needed, exact | Paths rounded up | Share once rounded up | Amp reading at 30 A |
|---|---|---|---|---|
| 30 percent | 1.3659 | 2 | 38.5567 percent | 11.5670 A |
| 40 percent | 2.1248 | 3 | 48.4875 percent | 14.5463 A |
| 50 percent | 3.1872 | 4 | 55.6548 percent | 16.6964 A |
So the 30 to 50 percent range in the exam diagrams is not describing one illegal green screw. It needs 2 to 4 parallel metallic paths, and once you round up it overshoots to 38.5567 through 55.6548 percent. That is the useful version of the claim: the number is a count of metal paths, not a property of the bond.
Why a single green screw is the wrong unit of analysis
Almost every ranking page frames the mistake as “somebody left the bonding screw in a subpanel.” That is true and incomplete. Remove the screw and you remove one connection, but the current was using every bonded path between the two panels. A site with a metal raceway, a shared cable tray and a bonded water line can still show current on metal after the screw comes out, because the second bond may be somewhere else entirely. The diagnostic is not a screw hunt. It is a clamp reading plus a list of parallel paths, checked against the divider you computed from the two conductor sizes.
The five legal moves and the one fatal edit
NEC 250.6 is a short section and it is worth reading in order. We used the NFPA 70 public comment reproduction, which prints (A) through (E) with the subsection text, and cross-checked it against a 2017 Mike Holt training sample, a DOE handbook restatement and two code commentaries. All of them carry the same qualifier in (B)(1). One exam prep page does not.
Nine clauses, four moves
| Clause | What the reproduction says | What a crew may do |
|---|---|---|
| 250.6(A) | Install and arrange grounding and bonding to prevent objectionable current | Fix the layout before energising |
| 250.6(B) | Four alterations are permitted, provided 250.4(A)(5) or (B)(4) are still met | Keep the effective ground fault path intact |
| 250.6(B)(1) | Discontinue one or more but not all of such connections | Remove the redundant bond, keep one |
| 250.6(B)(2) | Change the locations of the grounding or bonding connections | Move the bond to the one allowed point |
| 250.6(B)(3) | Interrupt the continuity of the conductor causing the current | Break the parallel metal path itself |
| 250.6(B)(4) | Take other suitable remedial and approved action | Get the AHJ to approve it first |
| 250.6(C) | Fault current and current from required bonds are not objectionable current | Find the fault, do not open the path |
| 250.6(D) | No electronic equipment on circuits without an EGC, and EMI is excluded | Do not lift a ground for a hum |
| 250.6(E) | A listed isolator may block cathodic protection DC while passing AC fault current | The one device allowed in series |
The opening line of (B) is the gate everyone skips. The four moves are permitted only if the effective ground fault current path requirements of 250.4(A)(5) or (B)(4) are still met. Read that as a condition on the whole list, not a footnote. You are allowed to change the topology of the grounding system. You are not allowed to end up with no fault path.
The three words that disappeared
Here is the NFPA reproduction: “Discontinue one or more but not all of such grounding or bonding connections.” Here is the exam prep paraphrase: “Disconnect one or more such grounding connections, for example remove illegal green bonding screws or bonding straps in downstream subpanels.” Three words are gone, and the verb changed from discontinue to disconnect. A DOE handbook restates the same intent in plain language: “At least one grounding connection must remain.”
A specialist site has an entire article about exactly this misreading. It describes an apprentice with a clamp meter on the ground bar, reading a couple of amps, concluding that 250.6 says the ground wire comes off. The article is right, and it still stops at the warning. Nobody prints what the missing three words are worth in fault current. So we will.
What an AHJ approved remedy can look like
The fourth permitted move is the one with room for engineering. A trade commentary gives a concrete example: bond two grounding connections together with a bonding jumper so they sit at the same potential and no current flows between them. That is a change in topology, not the removal of a safety path. It also notes that “isolated ground” receptacles are commonly installed with this idea in mind and are not what the section means. Either way, the move goes through the AHJ, and the effective fault path is checked afterward. Put that check in the commissioning record, with a clamp reading before and after.
The illegal fix, measured in fault current
Take the same published example that gave us 7.1848 A. It also gives a 120 V circuit, 0.386 ohm for the phase conductor plus the metal equipment grounding path, and 25 ohm for a ground rod and the earth. Those are published inputs, not our measurements, and they are enough to show what the missing qualifier costs.
310.8808 A against 4.8 A
| Quantity | Metal fault loop in place | Earth only, after the EGC is lifted | Ratio |
|---|---|---|---|
| Fault path impedance | 0.386 ohm | 25 ohm | 64.7668 times |
| Fault current at 120 V | 310.8808 A | 4.8000 A | 64.7668 times |
| In multiples of a 20 A breaker | 15.5440 times | 0.2400 times | 64.7668 times |
| Voltage on the metal case | Near zero | 120.0000 V | Full line to neutral |
| Published rounded pair | 311 A | 4.8 A | 64.7917 times |
A 0.386 ohm loop gives 310.8808 A, which is 15.5440 times a 20 A breaker rating and well into its instantaneous trip. A 25 ohm rod gives 4.8000 A, which is 0.2400 times that rating, or 76.00 percent below the point where the breaker would even begin to care. The case sits at essentially 120 V. The rounded pair on the published page gives 64.7917 times; our unrounded ratio is 64.7668 times. Both say the same thing.
Fifteen point five times down to zero point two four
The breaker is the whole reason the metal loop exists. With the equipment grounding conductor in place, a phase to case fault becomes a short circuit that clears in milliseconds. With it removed, the same fault becomes a 4.8 A load that never clears, and the enclosure stays energised. That is why 250.4(A)(5) says the earth shall not be used as the sole effective ground fault current path, and why 250.6(B) makes the four alterations conditional on that path still existing. The three missing words are not a drafting nicety. They are the difference between a trip and a live cabinet.
Why the earth is never the return path
A ground rod is for lightning, surges and holding the system at a reference potential. At 25 ohm it is not a fault return path for a 120 V circuit, and no amount of extra rods changes the arithmetic enough to matter here. If someone on site proposes “lifting the ground” to quiet a hum, the answer is not a code citation. It is 310.8808 A versus 4.8 A, written on the whiteboard before anybody opens a panel.
What the code excludes before you reach for the meter
The paragraph most often skipped is (C), and the one most often quoted backwards is (D). Together they decide whether a clamp reading is a wiring error, a live fault, or an IT complaint.
Fault current is not objectionable current
The reproduction says currents resulting from abnormal conditions such as ground faults, and currents resulting from required grounding and bonding connections, shall not be classified as objectionable current for the purposes of 250.6(A) and (B). So a clamp meter showing current on an EGC during a fault is showing the path doing its job. The correct response is to clear the fault, not to alter the grounding system. A 2017 training sample makes the same point with a different phrase: temporary current from a ground fault is certainly objectionable in plain English, but it is not objectionable current under the NEC definition.
EMI is not either
250.6(D) has two sentences and both are awkward for the ranking pages. The first says the section is not permission to run electronic equipment on circuits without an equipment grounding conductor. The second says currents that introduce electromagnetic interference or data errors in electronic equipment shall not be considered the objectionable currents addressed in this section. Ten of our ten leading pages list EMI or data errors among the hazards of a second bond. That is a real engineering consequence, and we compute its scale in the next section. It is just not the legal basis for opening a grounding path. If a server rack hums, the fix is a power quality study, not a pair of diagonals.
One listed device is allowed in series
250.6(E) permits a listed isolator in the equipment grounding conductor path when DC from a cathodic protection system has to be blocked. It has to pass AC ground fault current while blocking the DC. That is the narrow exception that proves the general rule: nothing else goes in series with the equipment grounding path. A commentary on the section spells out the implication, that otherwise no series device belongs there at all. So the tool kit for stopping stray neutral current is a change in topology, never an added impedance in the safety conductor.
A 200 A service, with the meter on it
The 100 ft example is a small feeder. Buyers asking about a main bonding jumper are usually standing in front of a service, so run the same divider on a 200 A service with a second bond downstream. Chapter 9 Table 8 gives 3/0 AWG stranded copper as 167,800 circular mils and 0.0766 ohm per 1,000 ft, and 6 AWG copper as 26,240 circular mils and 0.491 ohm per 1,000 ft. Table 250.122 gives 6 AWG copper as the equipment grounding conductor for a 200 A overcurrent device, a row confirmed across four published lookup tables.
Two conductor sizes, one more divider
| Item | Value | Where it comes from |
|---|---|---|
| Neutral conductor | 3/0 AWG copper, 167,800 cmil | Chapter 9 Table 8 |
| Equipment grounding conductor | 6 AWG copper, 26,240 cmil | Table 250.122, 200 A device |
| Neutral resistance at 100 ft | 0.007660 ohm | 0.0766 ohm per 1,000 ft |
| EGC resistance at 100 ft | 0.049100 ohm | 0.491 ohm per 1,000 ft |
| Resistance ratio | 6.4099 times | 0.049100 over 0.007660 |
| Split of 200 A by resistance | 26.9908 A | 13.4954 percent |
| Split of 200 A by area | 27.0460 A | 13.5230 percent |
The two methods differ by 0.0552 A here. On a bigger service the metal path is proportionally thinner, so the share falls from 23.9492 percent on the 100 A feeder to 13.4954 percent here. That is the opposite of the direction a hazard list would have you expect: the larger the service, the smaller the fraction, while the absolute current on the metal goes up. If a page says 30 to 50 percent here, it overstates by 2.2230 times at the low end and 3.7050 times at the high end.
26.9908 A that never trips anything
This is the part worth standing in front of the panel for. Nothing has failed. No breaker has moved. 26.9908 A flows on the equipment grounding conductor, the raceway and whatever else is bonded, all day, every day, because a load is running normally. The neutral carries the other 173.0092 A. A standard overcurrent device is looking at the phase current, not at where the return current chose to go, so it has nothing to react to.
A ground fault protection relay set near 1,200 A would not see 26.9908 A either, but that is a different device with a different job, and the resistance grounding page already covers it. What matters here is simpler: the metal that a person assumes is safe to touch is carrying a continuous load current, and the only way to find it is a clamp meter.
35.7696 W of heat, all day, all year
Current through a resistance makes heat, and this current never stops. At 0.049100 ohm, 26.9908 A squared gives 35.7696 W on that path. Over 8,760 hours that is 313.3418 kWh a year turned into warmth inside a raceway and at every coupling. Loose fittings are where that shows up first, which is why the fire hazard in the ranking pages is real even though their percentage is not.
| Quantity | Our divider | If 30 percent were true | If 50 percent were true |
|---|---|---|---|
| Current on the metal path | 26.9908 A | 60 A | 100 A |
| I squared R heat at 0.049100 ohm | 35.7696 W | 176.7600 W | 491.0000 W |
| Overstates real heat by | Baseline | 4.9416 times | 13.7267 times |
| Annual energy as heat | 313.3418 kWh | 1,548.4 kWh | 4,301.2 kWh |
| Share of neutral current | 13.4954 percent | 30 percent | 50 percent |
The heat column is where a wrong percentage becomes expensive. Heat goes as the square of the current, so a 2.2230 times overstatement of the share becomes a 4.9416 times overstatement of the watts, and a 3.7050 times overstatement becomes 13.7267 times. If someone sizes a corrective action from the printed range, they are solving a fire that is not there.
What the thresholds say
| Reference | Threshold | 26.9908 A as a multiple | What it means |
|---|---|---|---|
| Class A GFCI | 5 mA | 5,398.1677 times | It would trip, if its core enclosed this split |
| Common RCD | 30 mA | 899.6946 times | Same condition |
| Clamp meter at 0.1 mA resolution | 0.1 mA | 269,908.3862 times | Easily resolved |
| Clamp meter at 1 microamp resolution | 0.000001 A | 26,990,800 times | Easily resolved |
| The 30 A feeder against 5 mA | 5 mA | 1,436.9501 times | Even the small example is far above it |
| The 30 A feeder against 30 mA | 30 mA | 239.4917 times | Same |
That fourth column of the first two rows is doing real work. A GFCI or RCD only sees the difference between the conductors passing through its own sensing core. A split that happens on a feeder upstream of a downstream GFCI is invisible to that device, so “nuisance tripping” is not a guaranteed symptom. On the 100 ft feeder, a clamp around phase and neutral together reads 0 A, a clamp around the neutral alone reads 22.8152 A, and a clamp around the equipment grounding conductor alone reads 7.1848 A. Those three readings are the diagnosis, and the sum still comes back to 30 A.
The field around the loop, in millivolts
The ranking pages say a second bond causes electromagnetic interference, and 250.6(D) says EMI is not the objectionable current this section addresses. Both can be true. Here is the scale, computed from our own numbers and clearly labelled as our arithmetic rather than a measurement.
17.9939 microtesla at 0.3 m
Take a 30 m run with the return path 0.3 m away, so the loop encloses 9.0000 square metres. Using the long straight conductor approximation, the field at the loop is mu0 times I over 2 pi d, which reduces to 2 x 10 to the minus 7 times I over d.
| Current on the metal path | Field at 0.3 m | Flux through 9.0000 square metres | Induced emf at 60 Hz |
|---|---|---|---|
| 7.1848 A, the 100 ft feeder | 4.7898 microtesla | 43.1085 microweber | 16.2515 mV |
| 26.9908 A, the 200 A service | 17.9939 microtesla | 161.9450 microweber | 61.0518 mV |
| 200 A, if all of it went on metal | 133.3333 microtesla | 1,200.0000 microweber | 452.3893 mV |
The 200 A service is 3.7567 times the small feeder in both field and induced emf. The last row is 7.4099 times the 200 A service reading, and it is a ceiling rather than a real case. It shows why the numbers stay in millivolts: the diverted current is a fraction of the total, and the loop is a fraction of the building.
61.0518 mV at 60 Hz
The emf is 2 pi times 60 Hz times the flux, so 161.9450 microweber gives 61.0518 mV on the 200 A service. That is small compared with a 120 V circuit and large compared with a low level signal cable running in the same tray. Whether it matters depends on the equipment, the cable routing and the loop area, none of which a bonding page can know. That is exactly why 250.6(D) puts EMI outside the section: it is a power quality question, not a grounding permission slip.
Why this is a power quality study and not a code fix
Our field model is the simplest one there is. It assumes one long straight conductor, ignores the steel raceway, ignores the cancellation from other return paths and ignores harmonics. It is a scale estimate, not a survey. If a site has a real noise complaint, the answer is a power quality study and a cable routing change, with the effective ground fault path left exactly where the code put it. Write that in the commissioning notes so nobody reaches for a pair of diagonals later.
Eleven citations, three section numbers
Now for the part that makes a specification hard to write. We pulled eleven printed section references for this family of rules from the ranking pages and the code commentaries. They do not agree.
The table nobody prints
| Source | Section printed | What it is describing |
|---|---|---|
| Mike Holt 2017 training sample | 250.24(A)(5) | Load side prohibition |
| Electrical Flux | 250.24(A)(5) | Load side prohibition |
| Open Exam Prep, Kentucky | 250.24(A)(5) | Load side prohibition |
| EC&M, Grounding vs Bonding Part 3 | 250.24(A)(5) | Load side prohibition |
| Texas exam page, editions before 2026 | 250.24(A)(5) | Load side prohibition |
| Mike Holt 2023 newsletter | 250.24(B) | Load side prohibition |
| Texas exam page, 2026 edition | 250.24(B) | Load side prohibition |
| Electrical Contractor, Part 7 | 250.24(C) | Unspliced main bonding jumper |
| Voltage Lab | 250.24(C) | Unspliced main bonding jumper |
| PrepPass question bank | 250.24(C) | Grounded conductor sizing |
| Open Exam Prep, Kentucky | 250.24(C) | Unspliced main bonding jumper |
Five sources say (A)(5), two say (B), four say (C). Eleven citations, three section numbers, and two different rules mixed together: the load side prohibition and the requirement for an unspliced main bonding jumper. One SEO page runs a 2026 date in its title and a 2023 edition in its table header. A specialist exam page states the reason plainly: in editions before 2026 the load side rule was 250.24(A)(5), and in 2026 it is 250.24(B). We report that as a claim from a training source, not as a verified reading of the current NFPA text.
The edition line
Three dates are not the same date, and mixing them is how a submittal gets rejected. The edition a commentary was written against, the edition the local AHJ adopted, and the date a page was published can all differ by years. Our sources span a 2002 trade article, a 2017 training sample, a 2023 NEC newsletter, a trade magazine that says its figures are based on the 2023 edition, and a 2026 page with a 2023 table header. We did not verify the renumbering against the NFPA original. So put three fields on the drawing: the edition cited, the edition adopted locally, and the AHJ’s written ruling on which section number applies.
What to write in the submittal instead
Do not write “bond per 250.24(A)(5)” and leave it there. Write the physical requirement, which does not move with the numbering: one unspliced connection between the grounded conductor, the equipment grounding conductor and the service disconnect enclosure, at the service disconnect, with no neutral to case connection on the load side. Then cite the adopted edition and the section number the AHJ confirmed. The same habit applies to 250.6: name the clause letter, because (B)(1) and (D) say opposite things to somebody in a hurry.
Four jumper names and seven bond combinations
The other source of mix ups is vocabulary. Four different jumpers appear in Article 250, and they are not four names for one part. Our X0 page already prints the sizing table in full, so here we only separate where each one sits and which input decides its size. No gauge numbers are repeated.
Four names, four inputs
| Jumper | Sections named in the sources | Where it sits | What decides its size |
|---|---|---|---|
| Main bonding jumper, MBJ | 250.24(C) and 250.28 | Inside the service disconnect enclosure | Largest ungrounded service conductor |
| System bonding jumper, SBJ | 250.30(A)(1) | Any single point from the SDS source to the first disconnect | Derived phase conductors |
| Supply side bonding jumper, SSBJ | 250.102(C) | Supply side of the service or SDS overcurrent device | Supply side ungrounded conductors |
| Equipment bonding jumper, EBJ | 250.102(D) | Load side, across fittings and flexible connections | The overcurrent device rating |
Three of the four size from a conductor area. Only the fourth sizes from the overcurrent device, because on the load side the code knows the maximum current the fault path will be asked to carry. A code commentary adds that an external equipment bonding jumper is limited to 6 ft and must run with the raceway it bonds. The reason is impedance, and it is the same reason a short, clean fault path matters everywhere else in this article.
Seven rows, three legal
Now count the bond locations, not the jumper names. A service has exactly one allowed point. A separately derived system has two candidate points but only one may be used.
| First location | Second location | Bonds on that system | Verdict |
|---|---|---|---|
| Service MBJ | Subpanel bond | Two | Objectionable current |
| Service MBJ | No subpanel bond | One | Correct service |
| No service MBJ | No subpanel bond | Zero | No fault return path |
| SDS source SBJ | First disconnect SBJ | Two | Objectionable current |
| SDS source SBJ | No first disconnect SBJ | One | Correct SDS |
| No SDS source SBJ | First disconnect SBJ | One | Correct SDS |
| No SDS source SBJ | No first disconnect SBJ | Zero | No fault return path |
Three rows are legal and four are not. The service has one legal row out of three; the SDS has two out of four, because the code lets you choose either the source or the first disconnect. The two illegal rows with two bonds are the ones this whole article is about. The two rows with zero bonds are the ones the “just lift the ground” repair creates, and they are worse.
Where two bonds are actually allowed
A double ended service with a secondary tie is not two bonds on one system. A commentary on 250.24(A)(3) says you may use a single grounding electrode connection at the tie point of the grounded conductors from each power source when the dual feeds share an enclosure or a listed grouped arrangement. That is one connection at one tie point. Two genuinely separate separately derived systems also get one bond each, because each is its own source. Neither case puts two neutral to case connections on the same set of conductors, which is the only configuration that produces the parallel path we computed. If a drawing shows two, ask which system each one belongs to.
Six mistakes, in the order they get made
The keyword asks about mistakes, so here is the list in the order a job actually makes them. Notice that only one of the six is the extra bond itself. The other five are decisions made while describing it or fixing it.
The list, with a number attached to each
| Order | Mistake | What it actually changes | Number that catches it |
|---|---|---|---|
| 1 | Treating circulating current as a percentage | Replaces a divider with a guess | 23.9492 percent, not 30 to 50 |
| 2 | Counting one green screw instead of all bonded metal | Misses the other parallel paths | 8 paths give 71.5106 percent |
| 3 | Citing a section number without an edition | Three numbers are in circulation | 11 citations, split 5 / 2 / 4 |
| 4 | Treating EMI as the 250.6 reason | Reads (D) backwards | 61.0518 mV belongs to a power quality study |
| 5 | Lifting the equipment grounding conductor | Drops but not all from (B)(1) | 310.8808 A becomes 4.8000 A |
| 6 | Skipping the clamp reading after the repair | No proof that one bond remains | 0 A on phase plus neutral |
Why mistake five is the expensive one
The first four cost time and credibility. The fifth costs the fault loop. A crew that removes the last grounding connection has solved the hum and the meter reading, because there is no longer any metal path carrying current, and has also removed the only low impedance route a phase to case fault had. 310.8808 A becomes 4.8000 A, a 20 A breaker goes from 15.5440 times its rating to 0.2400 times, and the enclosure sits at 120 V. Nothing about that trade is visible from the clamp meter that started the whole conversation, which is why 250.6(B) puts 250.4(A)(5) in the opening line instead of at the end.
What a correct repair record looks like
Write four readings, not a paragraph. Before the change, clamp the equipment grounding conductor and record the amp value at a stated load. Clamp phase and neutral together and record 0 A. After removing exactly one redundant connection, repeat both readings. Then measure the fault loop impedance and record that it stayed low. Add the adopted edition, the clause letter you acted under, and the AHJ approval if you used (B)(4). That is a file an inspector can check and a future crew can trust.
What goes in the submittal
Our X0 page already has the fourteen line specification for the jumper itself, covering the sizing table, the single point and the eight subparts of 250.30(A). These twelve lines are different. They are about the current after the bond, and the drawing that proves where it goes.
Twelve lines, in the order a reviewer reads them
| Line | What to ask for | Why it is on the list |
|---|---|---|
| 1 | Adopted NEC edition and AHJ adoption date | Three section numbers are in circulation |
| 2 | Written AHJ ruling on which 250.24 subsection applies | Stops a numbering argument before it starts |
| 3 | One line diagram of every neutral to case connection | Seven bond combinations, only three legal |
| 4 | Neutral conductor size and circular mil area | First input to the divider |
| 5 | Equipment grounding conductor size and circular mil area | Second input to the divider |
| 6 | Every other bonded metallic path running in parallel | 2 to 4 paths are what 30 to 50 percent needs |
| 7 | Feeder length and reference temperature | Both cancel, so record them once and move on |
| 8 | Clamp reading on the EGC at a stated normal load | 26.9908 A on the 200 A service example |
| 9 | Clamp reading on phase and neutral together | Should be 0 A |
| 10 | Fault loop impedance before and after any alteration | 0.386 ohm against 25 ohm |
| 11 | Which single connection was removed and which one remains | The but not all requirement, written down |
| 12 | Power quality study reference in a separate file | 250.6(D) puts EMI outside this section |
Lines 4, 5 and 6 are the ones almost every quote leaves blank. Without them there is no divider to compute, and the conversation goes back to adjectives. If a supplier answers line 1 with a year and leaves lines 2, 6 and 10 empty, you do not have a comparable submittal yet.
Three fields that stop a numbering argument
Put the edition cited, the edition adopted locally, and the AHJ’s written ruling on the same drawing. Our eleven collected references span 2002, 2017, 2023 and 2026 publication dates, and the rule moved between subsection numbers in at least one of those accounts. A drawing that says “per NEC” with no edition is not a specification. A drawing that names the physical arrangement and then cites one confirmed number is.
Bring this to the site walk
The whole answer in one place
If you have five minutes in front of the panel, say this. The current on the metal path is set by the ratio of two conductor areas, not by a percentage from a hazard list. Length and temperature cancel, so ask for the two wire sizes. Count every bonded metal path in parallel, because that count is what turns 23.8825 percent into 38.5567, 48.4875 or 71.5106 percent. Then read 250.6(B) with its opening condition attached, keep one grounding connection, and never treat a hum as permission to lift a safety conductor.
The title of this article asks which mistakes cause circulating currents. The honest answer is that the extra bond is the small mistake. It creates a divider, and a divider can be measured and reversed. The larger mistake is a repair that removes the last connection, because it takes the same metal that was carrying 7.1848 A of nuisance current and makes it the only path a real fault could have used. 310.8808 A against 4.8000 A is the whole argument.
Eleven numbers worth keeping
| Number | What it is | Where it came from |
|---|---|---|
| 7.1848 A | Metal path current on the 100 ft feeder | 30 x 0.024500 / 0.102300 |
| 23.9492 percent | Its share of the 30 A | 7.1848 over 30 |
| 0.2782 percent | Agreement of the two independent methods | 7.1848 against 7.1648 |
| 31.3759 percent | Real 8 AWG area against 3 AWG | 16,510 over 52,620 |
| 2.3333 times | Area span behind 30 to 50 percent | 52,620 over 22,551.4286 |
| 71.5106 percent | Share with 8 parallel metal paths | 132,080 over 184,700 |
| 310.8808 A | Fault current with the metal loop in place | 120 over 0.386 |
| 4.8000 A | Fault current with earth only | 120 over 25 |
| 64.7668 times | What the wrong repair costs | 310.8808 over 4.8000 |
| 26.9908 A | Continuous current on a 200 A service | 200 x 0.007660 / 0.056760 |
| 35.7696 W | Heat that never stops | 26.9908 squared x 0.049100 |
Those eleven numbers fit on one page of a commissioning folder, and each one can be recomputed from the wire sizes on the drawing in front of you. Keep the pair 310.8808 A and 4.8000 A where everybody on the crew can see it.
If you have a one line diagram, two conductor sizes and a clamp reading from a finished job, send them over. We will tell you what share the divider predicts, how many parallel paths your number implies, which single connection can come out, and which of the twelve lines still need an answer before the file is signed off. Start here.
FAQ
How much current does a second bond actually put on the grounding path?
7.1848 A out of 30 A, or 23.9492 percent, on a published 100 ft example with a 3 AWG neutral and an 8 AWG equipment grounding conductor. The area method gives 7.1648 A, which is 0.2782 percent lower. A bigger 200 A service gives 26.9908 A, but only 13.4954 percent of its neutral current.
Why do so many guides say 30 to 50 percent?
Because they print a range without naming a conductor. To get 30 percent you need 22,551.4286 cmil of grounding path against a 52,620 cmil neutral, and to get 50 percent you need the full 52,620 cmil. That is a 2.3333 times area span, or 2 to 4 parallel metal paths in a real building.
Does feeder length change the split?
No. Doubling the 100 ft run to 200 ft doubles both resistances and returns exactly 7.1848 A again. Going from 75 C to 90 C scales both by 1.0485 and does the same. Length, temperature and copper resistivity all cancel. Only the ratio of the two conductor areas survives.
Can I just remove the downstream bonding screw?
Yes, if that is the redundant connection and one grounding connection remains. 250.6(B)(1) permits discontinuing one or more but not all of the connections, and the whole list in (B) is conditional on 250.4(A)(5) still being met. Record the before and after clamp readings and the fault loop impedance.
What do the three missing words actually cost?
310.8808 A against 4.8000 A, a 64.7668 times drop in fault current. A 20 A breaker sees 15.5440 times its rating with the metal loop and 0.2400 times with only a 25 ohm earth path. The metal case then sits near 120 V and the breaker never opens.
Is EMI a valid 250.6 reason to lift a ground?
No. 250.6(D) says currents that introduce electromagnetic interference or data errors are not the objectionable currents this section addresses. It also bars running electronic equipment on circuits without an EGC. On our 200 A example the loop induces 61.0518 mV at 60 Hz, which belongs in a power quality study.
Which NEC section do I cite for the load side rule?
It depends on the adopted edition. Eleven printed references we collected split 5, 2 and 4 across 250.24(A)(5), (B) and (C), and one training page says 2026 moved the load side rule from (A)(5) to (B). Name the physical arrangement, then cite the edition and the number your AHJ confirmed.
What is the difference between an MBJ and an SBJ?
The MBJ sits inside the service disconnect enclosure and is sized from the largest ungrounded service conductor. The SBJ serves a separately derived system and may sit at the source or at the first disconnect, but at one point only. Three of the four jumper types size from a conductor; only the EBJ sizes from an OCPD rating.
How many bond locations are legal?
Three rows out of seven. A service has one legal arrangement out of three, and a separately derived system has two out of four because either the source or the first disconnect may be chosen. Two bonds on one system create the parallel path; zero bonds leave no fault return path at all.
What should the clamp meter read?
0 A around phase and neutral together, 22.8152 A on the neutral alone, and 7.1848 A on the equipment grounding conductor alone on the 100 ft example. The three readings still sum to 30 A. On the 200 A service the EGC reads 26.9908 A at full neutral current.
How much heat does the stray current make?
35.7696 W continuously on the 200 A service, or 313.3418 kWh a year, using 0.049100 ohm for 100 ft of 6 AWG copper. If you believed 30 or 50 percent instead, you would compute 176.7600 W or 491.0000 W, overstating the heat by 4.9416 or 13.7267 times.
What should the RFQ ask for?
Twelve lines: adopted edition, written AHJ ruling, a one line diagram of every neutral to case connection, both conductor areas, a list of every parallel bonded metal path, length and temperature, two clamp readings, fault loop impedance before and after, which one connection stays, and a separate power quality study reference.

