How to Use Voltage Drop to Find a Bad Connection

Why this matters

A high-resistance connection is the fault that hides from every test a tech normally runs. It passes continuity. It reads full voltage with the load off. The breaker holds. Then it cooks its own lug for two years, browns the insulation, and either opens on a cold morning or starts a fire in a wall you never opened. The only thing that reveals it is current flowing through it, and the only measurement that sees it is the voltage across it while that current flows.

This is a five-second measurement once you know where to put the probes, and it converts "somewhere in this circuit" into a specific lug.

The steps below are ordered by what you lose if you skip them, not by the order your hands move. The first one is the only one that can kill you, so it is first.

Step 1: clear the energized-work gate and rate your instrument

Skipping this does not cost you a diagnosis, it costs you an arm or your life, which is why it outranks everything else here.

This measurement has to be taken live, because a dead circuit carries no current and a joint with no current through it drops no voltage. 29 CFR 1910.333(a)(1) permits energized work only where de-energizing would introduce additional or increased hazards or is infeasible due to equipment design or operational limitations, and diagnostic testing and troubleshooting is the recognized case that qualifies. Work inside that exception deliberately, not by default: if the panel can be shut down and the fault can still be found dead, shut it down.

In the same breath, the instrument. The meter and its leads and probes must be rated at or above the voltage at the point you are touching and at the right measurement category for that point (CAT III for distribution panels and fixed equipment, CAT IV at the service entrance). An under-rated or damaged-lead meter placed on a line-voltage fault is the failure that produces an arc flash across the probes and into your face. Inspect the leads for nicks before every use. Wear the arc-rated clothing, face protection and voltage-rated gloves your shop's arc-flash risk assessment calls for under NFPA 70E-2021, Article 130, and keep your other hand out of the enclosure.

When you move from measuring to repairing the joint, the gate closes: open the disconnect, lock and tag under 29 CFR 1910.333(b)(2) (29 CFR 1926.417 for construction work), and prove dead live-dead-live per NFPA 70E-2021, 120.5 before a wrench touches a lug.

Step 2: get the load running at its normal current

Skip this and you get a clean zero across a connection that is genuinely failing, write "no fault found," and hand the callback to whoever comes next. It is the most expensive diagnostic error in the whole procedure because it actively clears a real fault.

Ohm's law is doing the work here. A connection with 0.3 ohms of resistance drops 0.3 x 12 = 3.6 V at 12 A, and 0.3 x 0 = 0 V at no load. The resistance did not change; the current did. So put the equipment in its normal operating state - compressor running, heater energized, motor loaded - and take the reading there. If the load cycles, take the reading during the cycle, not between cycles.

Step 3: measure across the joint, never from the joint to ground

Skip this and the drop disappears into your reference point. Measuring line-to-neutral at both ends of a joint means reading, say, 121.0 V and 117.4 V and trying to see 3.6 V of difference between two numbers taken at different times with a meter that has its own tolerance. Put one probe on the conductor entering the connection and the other on the metal it lands on, and the meter reads the drop directly with nothing to subtract.

Same rule at a fuse, a switch, a contactor pole, a splice, a terminal strip and a breaker: probe across the device, not around it.

Step 4: clamp the current at the same moment

Skip this and you have a voltage with no meaning. Volts alone cannot be ranked, compared between circuits, or turned into a repair priority. Volts with amps gives you both numbers that matter: resistance (V / I) and heat (V x I).

Take both readings in the same minute, on the same conductor, with the same load state. A drop recorded at unknown current is a note, not a measurement.

Step 5: compare against an identical joint on the same circuit

Skip this and you have no baseline, so you either condemn a normal reading or excuse an abnormal one. The other pole, the other leg, or the neutral lug in the same panel is carrying similar current through a similar connection, and it is the fairest comparator you will ever get: same conductor material, same temperature, same meter, same instant.

State the gate this way, per connection, at the circuit's normal operating current, with the load running. Flag the connection if either is true:

• it drops more than roughly 10 times what an identical connection on the same circuit drops, or
• it drops more than about 100 mV in absolute terms, and repair before leaving site above about 0.5 V.

Those absolute numbers are a common starting point, not a code value. Tune them to your equipment: a large motor starter carrying 60 A has more legitimate drop across a pole than a 3 A control circuit does. The comparative half of the gate travels better than the absolute half, so lead with it when the two disagree.

Step 6: re-measure after the repair, at the same current

Skip this and you do not know whether you fixed the joint or merely disturbed it. A corroded lug that gets backed off and retightened often reads fine for a week because you scraped through the oxide, then returns. A re-measurement at the same load current tells you which happened, and it is your proof to the customer.

The worked case, carried through

A 120 V circuit feeding a piece of fixed equipment that "runs weak and the plug end gets warm." Load current, clamped: 12.0 A.

The whole-circuit read. At the breaker lug, 121.0 V line to neutral under load. At the equipment terminals, 113.8 V under load. Total drop is 121.0 - 113.8 = 7.2 V, which is 7.2 / 121.0 = just under 6 percent of the supply. For context, NEC Articles 210 and 215 carry informational notes recommending roughly 3 percent on a branch circuit and about 5 percent total from service to the farthest outlet; those notes are recommendations for conductor sizing, not enforceable limits and not a diagnostic threshold, so treat 6 percent as a signal that something in this circuit is eating volts, not as a violation.

The bisection. Rather than pulling the run apart, probe across each connection in turn with the load running. Across the equipment's line terminal lug: 3.6 V. Across the neutral lug two inches away, carrying the identical 12.0 A: 0.04 V.

That is a ratio of 3.6 / 0.04 = 90 times, which clears the comparative gate by an order of magnitude and the absolute gate by a factor of seven. No further searching is needed.

What the numbers say about it. Resistance at that joint is 3.6 / 12.0 = 0.30 ohms. Power dissipated in it is 3.6 x 12.0 = 43.2 W, all of it concentrated in a contact area about the size of a fingertip. That is why the housing is warm, why the insulation is discoloured, and why this joint was going to fail on its own schedule. It also accounts for 3.6 of the circuit's 7.2 V of loss, exactly half the total, from a single connection.

The repair. De-energized, locked and tagged, proved dead. Do not re-torque a lug that has already been running hot: the conductor strands and the lug have annealed and oxidized, and re-torquing a joint that has cooked buys weeks. Cut back to bright conductor, and replace the terminal or the device the terminal is part of. If the joint is aluminum to copper, use a connector listed for the combination with the compound the connector's instructions specify, because that pairing is a galvanic and thermal-expansion problem, not just a mechanical one.

The confirmation. Re-energized, same load, clamped again at 12.0 A. Across the repaired joint: 0.03 V, which is 0.36 W instead of 43.2 W. Equipment terminal voltage now reads 117.3 V, so total circuit drop is 3.7 V, or 3.1 percent. That is a hair over the 3 percent informational note and it is the right answer: what is left is conductor length on a long run, distributed along the wire rather than concentrated in one lug, and no single connection on the circuit now reads above 0.05 V.

Cross-checks that catch the cases this method misses

A thermal scan finds what the probe cannot reach. 43.2 W in a small connection is a hot spot a non-contact thermometer or an infrared camera sees through the cover. Use it to survey a panel you have no reason to open, then use voltage drop to confirm. Scan with the cover on where the equipment allows it, because opening a panel that contains a known failing connection is exactly when it lets go.

Drop that changes while you watch it is a loose connection moving under thermal expansion or vibration, not a fixed resistance. Note the range, not one number, and treat any joint whose reading wanders as a repair regardless of its lowest value.

Drop that appears on the neutral or grounded conductor matters as much as drop on the ungrounded conductor and gets missed because techs probe the hot side by habit. Probe both.

No drop anywhere and still low voltage at the load means the loss is upstream of your first probe point, at the service, or in the utility's conductors. Move your reference point back one device at a time rather than assuming the circuit is fine.

References

• 29 CFR 1910.333(a)(1) for the energized-work justification, and 29 CFR 1910.333(b)(2) with 29 CFR 1926.417 (construction) for de-energizing, lockout and tagging before repair
• NFPA 70E-2021, Article 130 for energized-work risk assessment and PPE, and NFPA 70E-2021, 120.5 for the live-dead-live proving sequence
• NEC Articles 210 and 215, informational notes on recommended voltage drop for branch circuits and feeders
• Manufacturer instructions for listed connectors and the compound required for aluminum-to-copper terminations
• See related: Ohm's Law as a Field Tool Rather Than a Formula; What a Resistance Reading Tells You and What It Hides