EER, SEER and CSPF: three scales that never line up
Three acronyms turn up on aircon listings, and none of them converts into another. Each was written by a separate standards body, against its own assumed year, in its own unit. A shopper reading them as one scale ends up ranking paperwork instead of machines.
By Team Snowflake | Updated 8 Aug 2026
What each of the three actually measures
The three acronyms grade one property under three separate rulebooks. Each divides the cooling a machine produces by the power drawn producing it. What separates them is how many operating conditions the figure covers, and which document fixed those conditions.
EER is the single-point member of the set. The United States federal test procedure defines it as the ratio of the average rate of space cooling delivered to the average rate of electrical energy consumed, established from one test or one interpolated set of operating conditions. A value produced under the current version of that procedure carries the label EER2.
SEER covers a season instead of a point. The same body of law defines it as the total heat removed from the conditioned space during the annual cooling season, in British thermal units, divided by the electricity consumed across that same season, in watt-hours. SEER2 is the version computed under the revised procedure used for compliance from January 2023.
CSPF is the international seasonal figure, defined in ISO 16358-1. That standard sets the annual heat the equipment can remove from indoor air in active mode against the annual energy it consumed doing so. How the seasonal weighting is assembled, and why part-load efficiency dominates it, sits on the page devoted to that metric.
None of this makes one figure a finer-grained version of another. A seasonal number is not a single-point number with more decimal places behind it. It is a separate quantity, produced by a separate procedure, describing a separate span of operation.
| The figure | What the measurement spans | The document that defines it |
|---|---|---|
| The figureEER and EER2 | What the measurement spansOne test at one set of operating conditions | The document that defines itUnited States federal test procedure, single-point method |
| The figureSEER and SEER2 | What the measurement spansA cooling season split into eight temperature bands | The document that defines itThe same federal procedure, seasonal method |
| The figureCSPF | What the measurement spansA cooling season assembled from outdoor temperature bins | The document that defines itISO 16358-1, an international test standard |
| The figureCOP and weighted COP | What the measurement spansOne test point, or two of them blended together | The document that defines itThe criteria the local requirement is written in |
The same three letters, two different unit bases
EER does not settle on one meaning either. Under the United States procedure the ratio is expressed in Btu per hour over watts, so a published value lands in the low double digits. The symbol table in ISO 16358-1 gives the same abbreviation in watts over watts, which carries no unit and sits in single figures.
European rules do it a third time. There a rated EER is the declared cooling capacity in kilowatts divided by the rated power input in kilowatts, which again produces a bare ratio rather than a Btu-based count.
Two sheets can therefore both print EER and mean numbers that share no scale. Nothing on either page announces the mismatch. A reader who takes the bigger value as the better machine has ranked two unit systems against each other.
A season is an assumption, and it belongs to a climate
A seasonal figure cannot be computed without first inventing a year. The procedure has to state which outdoor temperatures occur and what share of the running hours belongs to each, or there is nothing to average across.
The American cooling season is eight outdoor temperature bands carrying fixed weights. The coolest band sits at roughly 18 to 21 degrees Celsius and the hottest at roughly 38 to 40, and each is assigned a stated fraction of the season's hours.
Close to half of those hours fall in the two coolest bands. Together those two top out near 23 degrees. The Meteorological Service Singapore says local minimums usually stay above 23 to 25 degrees at night. Daytime maximums usually stay under 31 to 33. The bands covering that afternoon range hold about a sixth of the American season. The hottest band of all holds less than half of one percent.
A figure computed on that profile is therefore answering a question about a temperate year, where most cooling hours are mild ones. A machine working here holds the upper end of that distribution through most of the day and never reaches the bottom of it. The ranking is not false. It simply belongs to another climate.
The European framework exposes the same assumption from another angle. Its rules define one reference cooling season and three reference heating seasons, named average, warmer and colder. A regime that must specify three winters before it can grade a machine is not describing a year that resembles this one.
What the weighting does to a ranking
Weighting decides outcomes rather than presentation. Two machines that separate clearly at high outdoor temperature can finish level on a profile spending most of its hours somewhere cooler, because the conditions where they differ were handed a small share of the season.
This is the mechanism behind the instruction to compare like with like. It is not an appeal to tidiness. A figure computed on a foreign profile was asked a question about foreign weather, and the answer it returned is correct for that question and no other.
Which measure the local criteria are written in
None of the three carries the local requirement. In its November 2024 overview of the labelling and standards regime, NEA stated the efficiency requirement for household split-type air-conditioners in COP. Fixed-speed models answer to a full-load reading. Models that vary their output answer to a blended one.
CSPF does appear in that same table, and the column heading is the load-bearing part. It is presented as an equivalent. The requirement a model has to clear, and the figure a registration is granted against, is the COP one. What that ratio measures has its own guide, and so does the cooling seasonal performance factor beside it.
Neither EER nor SEER appears anywhere in it. They are not older measures or weaker ones. They belong to another jurisdiction's test procedure and were never part of the local paperwork, so no registration here produces them and no energy label carries them.
The practical consequence is narrow and worth stating plainly. A model can hold a strong SEER on an American listing and still have to be registered here on a measure that listing never quotes. Both facts can be true at once, and neither one predicts the other.
Where the tick bands sit in this
A tick rating groups models into bands on the local measure. It separates candidates only until two of them land in the same band. None of the three metrics here feeds it. Reading a foreign value as evidence of a band produces a claim nobody tested.
The floor beneath the lowest band comes from the minimum energy performance standards, which decide what may lawfully be supplied rather than how a model is graded. Both of those subjects have their own pages, and the numbers belong there rather than here.
Reading a shortlist where the metrics do not match
The working rule is short. Every candidate has to be quoted on the same measure, and nothing joins the comparison without its metric printed beside it. A value stripped of its label is not a rough guide to anything. It cannot be placed on a scale at all.
Following that rule is harder than it sounds, because the numbers arrive from different places. A showroom quotes what the local registration produced. A review quotes whatever its own market publishes. A brochure quotes whichever value flatters the model, and one model can have several to choose from.
Push back when a seller answers a question about the metric with a bare number. Asking which measure a figure sits on is not pedantry. Without that answer there is no telling whether two models were graded against the same thing, and a seller who cannot say has quoted a number rather than described a machine.
What follows sorts the shapes this problem takes, and what each one leaves unresolved.
- Ask for the metric name and the test standard together. Either one on its own still leaves the figure ambiguous.
- Insist that every candidate is graded on the same yardstick. A mixed list ranks documents rather than machines.
- Discard a value whose metric is not printed. There is no safe way to guess which of the three it belongs to.
- Never set a Btu-based count beside a bare ratio. Those are two scales, not two scores on one scale.
- Check the date and the market on any sheet handed to you. A revised metric returns a different number for an unchanged machine.
| What you are handed | Why it will not rank | What to ask for instead |
|---|---|---|
| What you are handedA bare value, metric unstated | Why it will not rankNothing on the page says which scale it sits on | What to ask for insteadThe metric name and the standard beside it |
| What you are handedEER from an American listing | Why it will not rankCounted in Btu per watt-hour, not as a bare ratio | What to ask for insteadThe value the local registration was granted on |
| What you are handedSEER from an overseas review | Why it will not rankAveraged over a season this climate never supplies | What to ask for insteadThe same local figure for every shortlisted model |
| What you are handedCSPF from a manufacturer sheet | Why it will not rankPublished as an equivalent, not as the requirement | What to ask for insteadThe measure the local bands are actually built on |
| What you are handedOne acronym, two countries | Why it will not rankThe unit basis differs, so the ranges differ | What to ask for insteadA single document quoting both models the same way |
A specification sheet built for another market
A sheet quoting a metric the local paperwork ignores is not a forgery. It was drawn up to answer a different regulator's question and it answers that question correctly. The trouble starts when it is carried into a comparison nobody designed it for.
Ask the supplier for the value the local registration produced. Where a model was registered here, that number exists, it was tested, and every rival on the shortlist can be quoted on the same basis. Where the supplier cannot reach it, the gap is worth noticing in its own right.
A missing local figure usually says something about the supply route. It rarely says anything about the machine. A unit brought in outside the authorised channel can be identical in every mechanical respect. It can still have no local registration behind it, and the guide on parallel import aircon covers what that changes.
Refuse the substitution that tends to follow. Asked for a local value and unable to produce one, a seller will often hand back the foreign figure with more conviction. Conviction is not a metric. The question stays open until a number arrives with the name of its measure attached.
What a revised metric does to an old number
A metric can be replaced while keeping most of its name, and the replacement does not return the same value. The American procedure moved from SEER to SEER2 for compliance from January 2023, and the newer figure comes out of a changed test method. An old listing and a current one can describe one machine and print two numbers.
So the year on a sheet matters as much as the market it was written for. A figure carrying no date and no metric name has two unknowns buried in it, and neither can be recovered from the number by itself.
Ready to get started?
Tell us what’s going on. Symptoms, setup, photos, anything we should know. We’ll assess and come back with the right next step.