Load Bank Testing for Emergency Generators: What’s Required
Load bank testing for emergency generators is required when the set can’t reach the code’s minimum load on the building’s own demand, and on a 36-month cycle for Level 1 systems. It isn’t required for every generator, every year, in every jurisdiction, and that’s the part most articles skip.
If you’ve read that the 2025 edition of NFPA 110 rewrote the load bank test, you’ve read something inaccurate. The 1.5-hour protocol at 50% and 75% has been in the standard for years.
What follows is what each clause actually asks for, when it applies, and what an inspector will accept as proof.
We build gensets from 8 kVA to 4,000 kVA and load test every one before dispatch. That means we spend most of our time on the difference between a test that happened and a test that proves anything. If you’re sizing a set and need the test report to satisfy the clause it cites, put the requirement in front of an engineer before the purchase order, not after the first survey.
Key Takeaways
- The annual load bank test is conditional. §8.4.2.4 applies only to installations that can’t meet §8.4.2.1.1 on building load: 30% of nameplate, or the manufacturer’s minimum exhaust gas temperature.
- The annual supplemental test runs 50% for 30 minutes and 75% for 60 minutes, 1.5 hours in total. That figure is in the 2022 edition, not new in 2025.
- Level 1 systems owe a separate test every 36 months at 30% of nameplate, for the class duration or four hours, whichever is shorter.
- The annual and triennial tests can be combined into one run: three hours at 30%, final hour at 75%.
- A report that says “test completed” proves nothing. The load level achieved at each step is the evidence.
When Load Bank Testing Is Required
The requirement isn’t a single rule. It’s four, and they trigger in different ways.
The monthly run and the 30 percent floor
Under NFPA 110 §8.4.2, a system in service has to be exercised monthly, and each run must last at least 30 minutes. §8.4.2.1.1 then sets what that exercise has to achieve. Loading has to maintain the manufacturer’s minimum exhaust gas temperature, or hold operating temperature at not less than 30% of the set’s standby nameplate kW.
That’s the floor everything else is built on. A diesel run for 30 minutes at 10% load satisfies the schedule and fails the substance. It’s also how wet stacking starts.
The clause most summaries leave out
§8.4.2.4 is the annual load bank requirement, and it begins with a condition. It applies to installations that do not meet §8.4.2.1.1, meaning sets that can’t reach the 30% floor or the exhaust temperature target on building load.
If your set holds that load every month, including on a permanent bank, the annual supplemental test isn’t owed. Where it can’t, the clause requires a supplemental annual run: half an hour at 50% of nameplate, then an hour at 75%, for at least 1.5 continuous hours.
Almost every vendor page states the test without the condition. “You need an annual load bank test” sells rentals and service contracts. “You might not” doesn’t, which is why the condition keeps disappearing.
This is an illustrative composite rather than a record of a specific client. A hospital engineering manager inherited a 500 kW set that had passed monthly exercises for six years. Every log said 30 minutes, running. Nobody had written down the load, and the building’s critical load sat at 14%. The set was compliant on paper and had never been asked to do the thing it exists for. When the incoming supply failed during a heat wave, the set picked up the load, ran hot, and tripped on high water temperature within twenty minutes.
The four obligations, side by side
| Obligation | Clause | Load | Duration | When it applies |
|---|---|---|---|---|
| Monthly exercise | §8.4.2 / §8.4.2.1.1 | ≥30% or manufacturer’s EGT | 30 minutes minimum | Every diesel genset in service |
| Annual supplemental test | §8.4.2.4 | 50% then 75% | 30 min + 60 min | Only where monthly load falls short |
| 36-month test | §8.4.9 | ≥30% or manufacturer’s EGT | Class duration, max 4 hours | Level 1 systems |
| Combined run | §8.4.9.7 | 30% then 75% | 3 hours + 1 hour | Where both tests fall due |
Section numbering is the 2022 edition, as printed in the NFPA public-input document for the 2025 cycle. The current edition is NFPA 110-2025, and older editions number these clauses differently, so quote the edition when you cite it.
Which sets this applies to
NFPA 110 applies to emergency power supply systems, and the level is assigned by the authority having jurisdiction under §4.4. Level 1 covers systems where failure could cause injury or loss of life: hospital critical branches, egress lighting in high-rises, fire pumps, some data centres. Level 2 covers the rest of what the code reaches.
A standby set installed for optional load, such as keeping a production line running through a utility interruption, may sit outside the standard altogether. That doesn’t make testing pointless. It makes it a business decision rather than a legal one.
Ask the AHJ in writing which level applies before you build a programme around it. An inspector won’t accept a vendor’s opinion of your classification, and neither will an insurer.
Three ways to reach the floor
There are only three: the building’s own load, a permanent load bank wired into the distribution, or a rented bank brought to site.
Building load is free and often unavailable. Hospitals, data centres and offices typically have less critical load than they assume, because the set was sized for growth.
A permanent bank costs capital and space, and pays back when you test often. Rental costs less up front and depends on a vendor’s calendar, which is a poor dependency on the week before a survey.
Whichever you choose, the number that matters is the same: can you hold 30% for 30 minutes, every month, with a record that shows it.
The 36-Month Load Bank Test for Level 1 Systems
Level 1 systems, where failure could cause injury or loss of life, carry a second obligation. Every 36 months the system must be tested for the duration of its assigned class, and where the class runs longer than four hours, the test may stop after four continuous hours. Diesel sets hold at least 30% of nameplate throughout, or maintain the manufacturer’s minimum exhaust gas temperature under §8.4.9.5.2.
This is the triennial generator test, and it catches a set that starts reliably and can’t run sustained. Starting is a battery and a starter. Running for four hours is cooling, fuel delivery, lubrication and everything the monthly half-hour never touches.
Duration comes from the class
Class is defined in hours in NFPA 110, and a Class 6 system is built to carry its load for six hours. The class duration is why the triennial test isn’t a fixed four hours for everyone: a Class 2 system and a Class 48 system aren’t asked for the same run. Type is the other half of the classification, and it’s measured in seconds, not hours.
Running both tests as one event
The annual supplemental test and the 36-month test can be combined. §8.4.9.7 sets the profile: three hours at 30% or more, then a final hour at 75% or more.
For a hospital or data centre already paying for a bank and a technician, that’s the difference between one mobilization every three years and two. It also produces a single record that answers both questions.
What the Test Is For Beyond Compliance
Wet stacking, in one paragraph
A diesel run at light load doesn’t reach combustion temperature, and unburned fuel plus moisture accumulates in the exhaust, turbo and combustion chambers. The set keeps starting. Then a real outage arrives, the engine is asked for 70% load, and the deposits restrict breathing. That’s wet stacking, and it’s one reason the code sets a 30% floor rather than a purely time-based rule.
What a load bank cannot prove
A bank proves the engine, alternator and cooling system deliver rated output under sustained load. It does not prove the transfer switch will transfer under a real outage, that the fuel will last the class duration, or that the battery will crank a cold engine in February.
Those are separate failures with separate tests, and a compliance file that only contains load bank reports is answering one question out of four. Remote monitoring covers some of the gap between tests, but it watches a set at rest. It can’t apply load either.
This is an illustrative composite rather than a record of a specific client. A cold-storage operator ran its 800 kW set for four hours every 36 months, on schedule, at whatever load the building happened to have. The building had 180 kW, 22.5% of nameplate. The engine ran clean every time at that load and failed the first summer outage at 55% load, because the radiator had never been asked to reject heat at that output. A load bank would have found it in year one. The calendar never did.
Running It Without Losing the Building
The automatic replacement rule
§8.4.2.3 is short and frequently broken. Loads used for testing have to be replaced automatically by the emergency loads if the normal supply fails. Manual switching doesn’t satisfy it.
In practice, the test has to be set up so the transfer equipment handles a genuine outage mid-test. Not a technician with a changeover routine and a clipboard.
Starting the test
§8.4.3 puts the start point at the transfer switch test function, or at the incoming breaker opened to mimic an outage. Pressing start on the controller tests the engine and skips the part that fails most often.
The sequence itself, with hold times and pass/fail limits, is a separate subject. This article is about what triggers it and what the paperwork has to show.
What an Inspector Will Accept as Load Bank Test Evidence
The record is the deliverable. A test that produces a line reading “test completed: yes” is not evidence of anything.
This is an illustrative composite rather than a record of a specific client. A contractor’s load bank test report reached a facility manager with the date, the duration and a tick box. The inspector asked for the load achieved. The answer was 22% of nameplate, which meant the annual test had never happened in the sense the clause means, and three years of records were reclassified as deficient.
A usable report contains, at minimum:
| Field | Why it matters |
|---|---|
| Load applied at each step, in kW and as % of nameplate | Proves the 50/75 profile was met |
| Duration at each step | Proves the hold times, not just the total |
| Voltage, frequency and current per phase | Shows the set wasn’t struggling |
| Coolant and oil pressure and temperature | Separates “ran” from “ran correctly” |
| Exhaust gas temperature | The §8.4.2.1.1 alternative to 30% load |
| Test initiation method | Shows §8.4.3 was followed |
| Instrument calibration status | An uncalibrated meter proves nothing |
Keep the reports well beyond the last survey. Inspectors ask for several years at once, and a gap in the file is read as a gap in testing. Most facilities put that filing duty inside a maintenance contract, which is also where the person who owns the record gets named.
Who signs the report?
In-house staff can run the monthly exercise. The annual and triennial tests need calibrated equipment and a technician who runs them often enough to notice when a reading is wrong.
Whoever performs it, every step needs a name and a time against it. Attach or reference the instrument calibration certificate. A report with no responsible person on it is a document, not a record.
Frequently Asked Questions
Does every emergency generator need load bank testing?
No. A set that holds 30% of nameplate or the manufacturer’s exhaust temperature on building load every month meets §8.4.2.1.1 and isn’t owed the annual supplemental test. Many commercial sets can’t clear that bar, which is why the practice is so widespread.
How long does the test take on site?
Budget half a day for the annual supplemental test. The loaded portion is 1.5 hours, and connection, warm-up and cool-down take the rest. The 36-month run occupies the class duration, up to four hours, and a combined run is four hours of loaded time before setup.
Can I combine the annual and 36-month tests?
Yes. The combined profile runs three hours at 30% or more, then one hour at 75% or more. It’s the cheapest way to satisfy both.
What does a failed test mean?
It means the set couldn’t hold the load, and the cause is usually fuel delivery, cooling or injection. Repair the fault, then repeat the same profile. Finding it on a scheduled test is cheaper than finding it during an outage, which is the whole point.
Is load bank testing required in every country?
No. NFPA 110 applies where adopted, which in practice means the United States and jurisdictions that reference it. Canada works to CSA C282, and other markets set their own rules.
U. S. military installation guidance is stricter. UFM 3-540-07 §8-3.2.2 calls for a one-hour monthly exercise, and its Method 2 holds no less than 50% of nameplate rather than NFPA’s 30%. A monthly run that misses either method triggers an extra semiannual test: 50% then 75%, two continuous hours, under the note to §8-3.2.3. RPIE POL/fuel systems owe that same run annually under §8-3.3.3 (UFM 3-540-07).
Does a resistive load bank satisfy the test?
For most diesel standby sets, yes. A resistive bank at unity power factor loads the engine, and the fuel, cooling and exhaust clauses are about the engine. Reactive or combined banks matter for verifying the alternator and voltage regulator at rated power factor. That’s a factory acceptance question, not a monthly compliance one.
My generator is Level 2. What changes?
The monthly exercise and the annual supplemental test still apply. The 36-month test is a Level 1 requirement. A Level 2 set still needs the load, and still fails the same way when it doesn’t get it.
Conclusion
Load bank testing for emergency generators comes down to one question: can this set reach the load the code sets, using the load it actually has?
If it can, the monthly exercise is the obligation and the annual test isn’t owed. If it can’t, §8.4.2.4 applies, and the 50/75 profile is the requirement. Add the 36-month test for Level 1 systems, and combine the two when both come due.
Two things decide whether the file stands up. The test has to start at the transfer switch, not the controller. And the record has to show load achieved, not just time elapsed.
That’s a lower bar than most vendors imply and a higher one than most facilities meet.
Specifying a set or reviewing a test report? Send us the rating, the critical load and the last report you have, and an engineer will tell you whether the documented test meets the clause it claims to satisfy.
