What is NEC 220.87?

What it is for

Somebody wants to add load to an existing service, and the question is whether the service can carry it. The standard answer comes from the Article 220 calculation, which adds up every load in the facility with the demand factors the code allows. On an operating facility that figure runs high, because it assumes far more of the equipment runs at once than ever does.

NEC 220.87 allows the existing load to be established from evidence instead: what the service actually drew. On most operating facilities the measured figure is well below the calculated one, which is the entire reason the section exists and the entire reason a load study is worth commissioning.

Path 1: a year of utility demand data

If the utility has had a demand meter on the service for more than a year, the maximum demand over those 12 months is the existing load. No recording is needed. This is the primary path, and it should be asked about first every time, because many commercial accounts hold interval data the account holder has never opened, and proposing 30 days of metering to a facility that already has a year of records is selling work that is not needed.

A year of kilowatt-hour bills is not a year of demand data. Consumption tells you how much energy was used over a month. Demand is the peak rate at which it was drawn, and that is what the section wants. An account billed on energy alone has no usable record for this path.

Path 2: the 30-day recording

Where a year of demand data is not available, the section allows the existing load to be recorded. The recording has conditions, and every one of them is a way a study can be invalid. It has to run for at least 30 days, continuously: a window with gaps is not a shorter valid window, it is an invalid one. The maximum demand is the highest average over a 15-minute interval, so a 1-minute spike is not the figure. It has to be taken on the highest loaded phase, or on all three with the highest used. It has to be taken while the facility is occupied and operating, because 30 days over a shutdown answers nothing.

And it has to include the larger of the heating or cooling load, and any other periodic or seasonal load, either by measurement or by calculation. That last condition is the one most often missed.

The seasonal trap

A 30-day window recorded in spring or autumn can capture neither the heating peak nor the cooling peak. The section requires the larger of the two to be included, so a shoulder-season recording is not automatically disqualified, but the report has to add the missing heating or cooling load by calculation and say that it did. Presenting a shoulder-season measured peak as the maximum demand, with nothing added, produces a figure lower than the truth, and the whole point of the study is that somebody sizes a service on that number.

Where the schedule allows it, record in the season that produces the peak. In a cooling-dominated facility that is summer. Heating load has to be measured in winter.

Two things that rule out the 30-day path entirely

The recording path does not apply where the service or feeder has a renewable energy system on it, or employs any form of peak load shaving. In both cases the meter is reading net rather than load: a site with solar shows a lower demand than the facility actually draws, because the array was covering part of it, and a service sized on that number is undersized on the first cloudy afternoon. A facility with solar or with a battery on demand management needs the utility data path, or a measurement on the load side of the generation, or the generation added back by calculation. The section does not allow the recording to be run and hoped over.

The arithmetic, and where the 125% goes

Once there is a valid maximum demand: the existing demand at 125%, plus the new load, must not exceed the ampacity of the feeder or the rating of the service. The 125% is applied to the existing measured demand. It is not applied to the new load, which is calculated the normal way under Article 220 because it is not installed yet.

There is a second, unrelated 125% in the code, and confusing the two produces a wrong number. Under the 2023 edition and earlier, a continuous load is counted at 125% of its rating when sizing circuits, and that rule often applies to the new load as well. It is a different rule. On a 400 A service with a 300 A measured peak and 40 A of new continuous load, the method gives 300 at 125% plus 40, which is 415 A and does not fit. Putting the 125% on the wrong term gives 300 plus 40 at 125%, which is 350 A and appears to fit. The second answer is wrong in the expensive direction: it says yes where the code says no, and it does so whenever the existing demand is the larger part of the sum, which is nearly always.

The section also requires that the feeder has overcurrent protection per 240.4 and the service has overload protection per 230.90. Those are conditions of using the method, not part of the arithmetic, and a reviewer will ask about them.

In the 2026 edition it is 120.87, and the arithmetic changes

The 2026 edition of the NEC moves Article 220 into a new Chapter 1 as Article 120, and the section becomes 120.87, Existing Loads and Equipment Replacement. The method did not move. The number did, and that makes the citation edition-dependent, which is unusual. Cite the number from the edition the jurisdiction has adopted, and only that number. Writing both on a submittal reads as not knowing which edition applies, which is exactly what it means.

Two things change in the 2026 arithmetic, as summarised by the secondary sources on the new edition. Load being removed from the service can be credited against the new load, which matters in the customer's favour: a contractor taking out old equipment while adding chargers has more headroom under 2026 than the same job scores under 2023. And the 125% on continuous loads comes out of the load calculation itself, remaining only in conductor and overcurrent device sizing. Where the adopted edition is not on record, the conservative computation is the 2023 one, without the removed-load credit, because it is the one that will not be rejected.

Adoption is state by state and lags publication by years. Massachusetts adopted the 2026 edition as 527 CMR 12.00, effective 24 April 2026, and there the edition that applies to a job vests with the electrical permit application date. Most jurisdictions will be on 2023 or earlier for some time. Assume nothing, and check the adopted edition before citing a number.

What a reviewer will ask for

The applicable edition, and therefore the article number and the formula. The feeder or service being evaluated, its rating, and where that figure came from. The existing overcurrent and overload protection. Which path was used, and if the recording path, why the utility data path was not available. The demand record itself, with its start, end and continuity, the 15-minute sampling interval, the metering method and where it was connected, and which phase. A study that cannot produce every item on that list is a study that gets sent back.