How we do this

The four certified figures

The certified data gives four capacity figures for each pairing: heating at 5°F, 17°F, and 47°F, and cooling at 95°F (cooling below). These are separate laboratory tests under the Department of Energy test procedure. A variable-speed unit may use different compressor speeds at different temperatures. A single-stage compressor has one speed; a two-stage unit's full-load figures use its high stage.

Single-stage test names above apply to fixed heating airflow. Appendix M1 also defines H1₂, H3₂, and H4₂ for single-speed compressors with qualifying variable-airflow or multiple indoor blowers. The compressor label in the certified data does not identify that distinction.

For variable-speed equipment, H1N uses the full speed the controls would use in normal operation at 47°F. The colder tests use the speeds defined for their own temperatures. These need not be identical, and the 47°F rating is not a measure of output at every available speed.

Test names and operating conditions: DOE Appendix M1, sections 3.6.1–3.6.4 and Tables 11–14.

The share is heat output at 5°F divided by the 47°F rating. Tests use different compressor speeds.

Capacities are certified by the manufacturer under the Department of Energy test procedure and published in EPA's certified data. This site publishes them as certified and adjusts nothing.

The 5°F test is optional under federal law and required by Energy Star for every certified heat pump, so almost every unit here carries a 5°F figure. The cold-climate mark needs a COP at 5°F of at least 1.75 and a 5°F capacity of at least 70% of the 47°F nominal capacity, verified in a separate controls check whose measured 5°F capacity can differ from the published one. The mark is EPA's, and this site reports it rather than recomputing it.

EPA publishes no field description for the three capacity fields. The 47 and 5°F figures are named in the Energy Star specification. The 17°F figure is named nowhere: H3 is the only 17°F heating test in the procedure, and this site reads the field as that test's result. That reading is an inference, and it is the one place on this site where the basis of a figure is not stated by its publisher.

Plain terms

COP

Heat delivered for each unit of electricity used. A COP of 2.5 gives 2.5 times the heat of an electric baseboard heater.

HSPF2

Heating efficiency over a typical season. Higher uses less electricity.

SEER2

Cooling efficiency over a typical season. Higher uses less electricity.

EER2

Cooling output divided by electrical input in the 95°F full-load test, in Btu/h per watt.

Pages other than this one and the data page name the heating tests by temperature and speed, not by test code: "At 47°F (normal speed)" is H1N for variable-speed equipment (H1 at high stage for two-stage equipment, and H1 at the one speed of a single-stage unit); "At 17°F (full speed)" is H3; "At 5°F (full speed)" is H4.

Between and below the test temperatures

A unit page draws each certified test as its own bar at its own outdoor temperature: 5, 17, and 47°F for heating and 95°F for cooling. Nothing is drawn between them. DOE's seasonal rating calculations interpolate in a straight line between the 17 and 5°F results when the 5°F test is available; the chart does not draw that line.

For variable-speed equipment, see Appendix M1 section 4.2.4.d. Single-speed and two-capacity calculations are in sections 4.2.1–4.2.3.

Power use at each temperature

A unit page gives three power figures for the pairing on screen. Each is worked out from certified fields, and none is joined to another: they sit on three different bases.

• Electrical input in the 5°F full-speed test: certified output ÷ (COP × 3,412), in kW.
• Electrical input in the 95°F cooling test: cooling capacity ÷ EER2, in kW.
• Seasonal average heating COP: HSPF2 ÷ 3.412, modeled for DOE climate Region IV.

They exclude backup resistance heat.

The median pairing

An outdoor unit is certified with indoor units, and the certified figures belong to the pair. One outdoor unit can carry thousands of certified combinations, and their figures differ.

A unit page leads with the median across that unit's own pairings, prints the full range beside it, and lets you select the pairing on your quote. The median describes the middle of the published pairings; it can change when pairings are added or removed. A build check re-derives every headline figure from the certified data and fails if a page prints anything else.

Cooling in the certified data

The certified data has four cooling fields for each pairing: the cooling capacity in the 95°F full-load test, EER2 in that same test, SEER2 over a modeled cooling season, and a yes-or-no field for peak cooling requirements. It has no cooling figure at any other outdoor temperature. This site prints the first three.

A size class is the nominal cooling capacity. Counted one value per unit, the median ratio of heating at 47°F to cooling at 95°F is 1.02. Heating at 47°F is 20% or more above cooling for 489 of the 8,418 units with both figures.

Size classes and comparison groups

Size classes are the nominal sizes the market sells: 6, 9, 12, 15, 18, 24, 30, 36, 42, 48, 54, and 60 thousand Btu. A unit is filed in the class its certified combinations most often fall in, by cooling capacity, computed separately for each buyer-facing type. Ties go to the smaller class, so a unit never drifts upward. A unit whose combinations span more than one class is listed in the class it is most often certified at and cross-listed, flagged, in the others it reaches. A size list ranks the units filed in the class first and the cross-listed ones after them.

A uniform 6,000-Btu grid was the first rule, and it filed the largest mini-split class at 9,000 Btu as 12,000, so it was replaced.

Comparisons are made within one size class and one type, one value per unit. The comparison group is a size class crossed with the type facet, mini-split or ducted, and every distribution in it counts each outdoor unit once, using that unit's own median pairing. Counting certified records instead would let a unit certified against hundreds of indoor pairings outvote one certified against three, and it did: the published 18,000 Btu median came out below the 12,000 Btu one, so a 1.5-ton unit read as smaller than a 1-ton unit.

A group needs at least 30 units before it is used as a comparison group. Below that it falls back to the size class across types, and every label that names it says so: "all 1.5-ton models" rather than "1.5-ton ducted models".

"Size" on this site is the nominal cooling capacity, in tons or Btu; a size class is named by it. "Capacity" is reserved for a certified output.

Square feet on a size page

A size page gives one rule-of-thumb range, for a typical home in a mild climate. It draws on design loads for metered homes in Efficiency Maine's 2026 study and modeled loads in NREL ResStock: 16 to 26 Btu/h per sq ft. We divide the median certified heating at 47°F of the size class (or of the size and type on a type page), one value per unit, each counted in its main size class, by that load per square foot and round to the nearest 50 sq ft. The nominal size is a cooling capacity, so it is not the divisor. Colder winters and drafty homes need more; the size calculator works from your home.

Example heating load

The dashed orange line on a unit chart is an example heating load: what a house that needs heat would ask of the unit as the temperature falls. It is not your house, and it is not a load calculation.

The 18 to 30 band comes from two published sources, and the site names both: Efficiency Maine's 2026 assessment of 78 metered Maine homes, whose design heat loss averaged 20.3 Btu/h per square foot at 5°F with a range of 11 to 32; and NREL's ResStock model of the US housing stock, which gives the spread by climate zone and vintage. An older, un-retrofitted house can need roughly double the rate, and a recent or deeply retrofitted one roughly half. Vintage moves that number more than climate does.

The bar chart on a unit page also draws an example cooling load for the same area: 12 Btu/h per square foot on a 95°F day, which is 1,000 square feet per ton, rounded to the nearest 500 Btu/h, falling in a straight line to zero at 65°F. 65°F is the zero-load temperature the federal rating procedure uses for cooling, so the heating and cooling lines meet there. The 12 comes from modeled and calculated loads for code-built homes where winters reach 5°F; nothing measured it. A house in a hot climate or with large west-facing windows can need about twice as much.

Size calculator

You give a ZIP or postal code, floor area, and a description of the house. The calculator uses a nearby suitable weather station to choose a climate band. It multiplies floor area by a field-based heating-load range for that band and the selected house condition. Age, insulation, windows, and draftiness suggest a condition that you can change. It estimates a cooling load the same way, from a modeled rate for your climate zone and the same house condition.

The station design temperature is the 1st percentile of hourly observations from 2015 to 2024. A station needs eight qualifying years with at least 7,000 readings each. Degree days and temperature-bin hours come from the same record.

The estimate uses the field-based range shown with its sources. The separate NREL ResStock distribution supplies modeled context by climate zone and vintage; it is not the rate applied to your floor area. Within a climate band, the estimate is not rescaled for every degree of local design temperature. Neither approach replaces a room-by-room load calculation.

Reading an estimate

• This is not a Manual J load calculation, the room-by-room method contractors use to size equipment. It does not replace one. CSA F280 is the Canadian equivalent.
• It is a range, never a single number. The range is the span of design heating loads that field studies measured in homes of that condition and climate, per square foot of heated floor area.
• The house condition moves this answer more than the climate, the floor area, or anything else on the form. It is pre-selected from your answers by the rule below, you can change it, and the result shows all three conditions side by side so you can see what the choice is worth.
• Your design temperature comes from one weather station's hourly observations, 2015 to 2024. It is good to about ±3°F, and to about ±8°F one time in twenty. The answer prints the station and its distance, so you can judge whether it speaks for you.
• Ten years is a shorter window than the thirty a published design table uses. It buys much denser station coverage, so the station is nearer.
• A ZIP code is a postal route, not an area. We answer with the Census ZCTA of the same digits. Where there is none, we answer from the middle of that ZIP3 and the answer says so.
• In Canada the answer is for a Forward Sortation Area, the first three characters of the postal code. A rural one, with 0 as its second character, can be the size of a small country, and the answer carries a warning when it is.
• Certified capacity belongs to a pairing of an outdoor unit with an indoor one. The pairing on your quote decides the real number.
• There is no certified figure below 5°F in the certified data, and nothing here extrapolates one. Where your design temperature is below 5°F, the answer says how many hours a typical year spends below it at your station, from the same hourly temperature bins, so you can see how much backup heat that means.
• There is no certified cooling figure above 95°F in the certified data either. Where your hot-day design temperature is above 95°F, the answer says so.
• The cooling estimate is modeled, not measured. It is the less certain of the two.

House condition rules

The condition is pre-selected by this rule, read in order: the first line that matches decides. You can change it afterwards, and the answer shows all three conditions either way.

The year it was built sets the starting point — before 1980 is Typical, 1980 to 1999 is Typical, 2000 or later is Tight, well insulated — and the three envelope answers decide the rest. A house from before 1980 that has been insulated and has double glazing is a typical house, not a drafty one.

Answer effects

The same rate per square foot is used for the one-room mode. One room inside a heated house has fewer exterior surfaces than a whole house of the same floor area, so the same rate per square foot is an upper bound.

Measured houses, for comparison

Every dataset behind the calculator's rates per square foot. Each carries its own definition of a design load, and the last row is a measured peak draw rather than a design load, which is why it is so much lower.

The modeled distribution beside them is NREL ResStock, 246 cells of design heating load per square foot by climate zone, vintage, and foundation. Its cells for homes built before 1980 run about 1.5 times higher than the measured homes above, so the field figures are the middle of the distribution and the modeled ones are the upper tail. The calculator multiplies floor area by the field ranges for that reason.

How the cooling estimate works

The cooling estimate multiplies floor area by a rate for your climate zone and house condition. The rates come from NREL ResStock, a model of the US housing stock: the peak cooling load each modeled home reaches in a typical weather year at its own location, per square foot, for single-family homes cooled whole by central air conditioning. The house condition picks the age of the modeled homes: drafty is homes built before 1980, typical is 1980 to 1999, and tight is 2000 or later.

It uses the modeled peak load, not the air conditioner size the model chose. The model's sizing step can add up to 15 to 30% to the load.

The range runs from the 10th to the 75th percentile of the modeled homes. The low end is wider than the middle half because the model runs high for new houses in hot, humid places: a house built to the 2009 code in Orlando, calculated room by room, comes to 9.3 Btu/h per square foot, and 40 room-by-room calculations in hot and mixed climates average 8.4. The model's median for homes built since 2000 in Orlando's zone is 14.7. Its 10th percentile is 9.7.

The rate is read by climate zone, not scaled by the hot-day temperature, the same way the heating rate is read by climate band. The zone's moisture letter counts: a dry zone (B) comes out at or above the humid zone (A) of the same number, except in zones 3 and 7. A Canadian postal area has no published zone, so one is derived from its heating degree days, and the answer says so.

Btu/h per square foot of conditioned floor area. A published zone with no row of its own (7, 8, and 5C) takes the nearest row with the same number, or the next zone down.

Heating load from fuel bills

A reader with last winter's bills can add a second heating estimate. The year's fuel, times the energy in one unit of it, times the efficiency of the equipment, is the heat delivered to the house that year. Where the same fuel also heats water or cooks, twelve times one summer month is taken out first. That heat, divided by the station's degree-days below a balance point, is the house's heat loss per degree. The heat loss per degree, times the gap between the balance point and the design temperature, is the heating load.

The balance point is the outdoor temperature below which the furnace runs. It is not on a bill, so the load is worked at 55, 60, and 65°F and the range is the span. A balance point with fewer than 500 degree-days at the station is left out, and so is one within 10°F of the design temperature.

The degree-days here are summed from the station's hourly temperature bins, so they differ from the degree-day figure printed with the design temperature, which uses daily means. In a hot place the two can differ by a third. A station's bins count only the hours it recorded, so the sum is scaled up to a full year where it has gaps.

• This measures the house as it was heated: the thermostat settings, the duct losses, and the heat from people and appliances are all in the bills. A room-by-room load calculation takes no credit for that heat, so the gap is measured from the balance point and not from the indoor temperature.
• The bills are one winter. The station figures are ten. A winter 10% milder than those figures makes this answer 10% low; a colder one makes it high.
• If the ducts run through an unheated attic or crawl space, some of the heat never reached the rooms. A ducted heat pump on the same ducts loses it too, so the estimate keeps it.
• How far this method sits from a Manual J is not established, so it is shown beside the floor-area estimate, not instead of it.

From the U.S. Energy Information Administration's Monthly Energy Review, Appendix A. The metric units are converted here from the US figures.

The reader can change the efficiency to the nameplate AFUE. The bands are from the U.S. Department of Energy.

Climate figures

• The design temperature is a percentile of the hourly temperatures recorded at the nearest suitable weather station from 2015 to 2024. It is computed from those observations.
• It describes the station, not your house. The distance and the elevation difference are published with the answer. A station 40 km away, or 300 m lower, is describing somewhere else's winter.
• A Canadian building department states the January 2.5% or 1% dry bulb. That is a different statistic from the annual value the calculator sizes on, so both are printed and each is named.

Canada

Canada is keyed on the Forward Sortation Area, the first three characters of a postal code, and that is not a choice: the file that used to map six characters to geography is discontinued and now licensed commercially, so three characters is as fine as an open source gets.

Calculator limits

Canadian limits

Rounding and display

Every share and count on a unit page is per certified combination of that unit unless it says otherwise. A share outside the 40–130% range, which covers more than 99% of the rows in the certified data, is published unchanged.

Raw values are never rounded: the JSON twin and the CSV carry the full figure. A share just below 100 prints as 99, not 100, so a unit short of its full rating is never shown as delivering all of it. Certified capacities are declared to the nearest 100, 200, or 500 Btu/h, so a tenth of a point in their ratio would be finer than the figures behind it.

The flag vocabulary

An empty cell is never a zero and never a blank. Where the certified data carries nothing, the cell carries a flag that says which kind of nothing it is.

Unit grouping

There is one page per physical outdoor unit, not one per badge. The certified data is keyed on brand × model string × product type, which puts the same machine on as many rows as it has badges. A page keyed that way would split one unit's pairings across several addresses and repeat every number.

Grouping decides which page shows a listing. It never changes a number, and every listing keeps its own row and its own line under "listed as". A trailing wildcard in a model number stands for the suffixes of one family and is shown on the family's page; an accessory bundle or an indoor/outdoor pair string is listed with its base unit.

On a unit page, pairings are grouped by indoor-unit class. The class is our inference from the model number, not a field in the certified data: a listed furnace means furnace-paired; mini-split indoor units form their own group. Central units with an indoor model and no listed furnace are grouped as air handlers. The certified data may not distinguish an air handler from a separate coil.

Identical certified values

It means exactly this: two records in the certified data carry the same certified numbers. Where a unit is listed under more than one brand, or where two different model numbers carry the same values, the site says so and shows the comparison.

The brands on the home page are in order of US search interest, measured in Sep 2026; scores are rounded to the nearest five before ordering, and then the brand with more units comes first. Where several units carry identical certified values, the main listing is a full model number before a wildcard pattern, then the most-searched brand, by the same rule, then the unit with the most pairings.

It does not mean the machines are the same, that they come off one line, or that they are interchangeable. The certified data has no manufacturer field, and this site never asserts one. It is an observation about the data, and it is published with what the data cannot show: the factory, the warranty, the dealer network, the parts, and the price.

Which brands are the same?

Known data-quality issues

These are properties of the certified data. The site reports them rather than correcting them, because correcting a certified value is not something a reader could check.

Change rules

The certified data moves constantly, and most of the movement is re-dating: one week moved 55,378 rows, 19% of it. Treating that as change would put a dishonest last-modified date on hundreds of pages. So the watched fields are split in two.

Changes to our method are a version, below.

Updates

A page's Updated date is the day its figures or content last changed. The certified data is checked daily, and a page keeps its date until a check finds a difference in what it shows. A unit page also carries the certification date the certified data gives for the pairing on screen, so a reader can tell a figure certified years ago from a recent one.

Updated Oct 26, 2026. Checked daily, last difference found Oct 24, 2026.

Method version

This is Method v1.0. The version is stated here only; every page carries the date its figures or content last changed.

A change to how a figure is derived, grouped, or displayed increments the version, and the change is recorded. A correction to a single page does not: that is a correction, and it goes in the corrections log.

Corrections

If a figure here disagrees with the certified data, the certified data is right and we have an error: tell us, and we will fix it and say what we fixed. If a figure here agrees with the certified data and you believe it wrong, the page says so and links the evidence, and we will not silently edit a certified value.

How to send a correction.