What Excess Air Does to Efficiency

Why this matters

Every combustion appliance in the field runs with more air than the fuel needs, and that surplus is not waste in the sloppy sense. It is insurance, paid for in heat that leaves up the flue. Techs get this backwards in both directions: some never touch it and leave a quarter of the stack loss on the table for years, and some chase the lowest oxygen reading they can achieve on a mild afternoon and hand back an appliance that makes carbon monoxide on the coldest night. The optimum is not the minimum. It is the smallest surplus that still holds carbon monoxide down under the worst condition the appliance will actually see, and that condition is almost never the one you are standing in.

Before you touch a single adjustment

Have a carbon monoxide monitor running in the space before the appliance fires, not after. OSHA's general industry permissible exposure limit for carbon monoxide is 50 ppm as an 8-hour time-weighted average at 29 CFR 1910.1000 Table Z-1, and NIOSH publishes a 200 ppm ceiling. If ambient carbon monoxide climbs toward those numbers, or anyone gets a headache, everyone leaves the building immediately, nobody touches a switch on the way out, and the space is ventilated from outside before anyone re-enters.

If you smell gas at any point, that is not a reading to take: everyone out immediately, no switches, no lights, no phone used inside, shut off at the exterior shutoff only if it is on your way out, and call from outside.

Four more that belong to this specific job. Do not reduce excess air on any appliance until you have first confirmed the heat exchanger is sound and the draft is correct, because leaning out an appliance with a defect converts a small carbon monoxide problem into a large one. Do not stand square in front of an open sight port while adjusting a firing burner, since a burner driven rich can roll flame out of the opening; work from the side and wear eye protection. And if you need a flue sampling port, drill only metal vent with eye protection against chips, never a plastic vent on a condensing appliance, which is a listed sealed pressurised system that must use the manufacturer's own test port. Let a sample probe cool before you handle it, or hold it with gloves rated for contact heat.

What the surplus is actually buying

Air and fuel do not mix perfectly in any real burner. There are always local pockets running rich, and a rich pocket cannot finish its reaction no matter what the room average is. Extra air is how you make sure the lean pockets outnumber the rich ones. The mechanism behind that, and the stoichiometric numbers it departs from, are covered in the combustion fundamentals article and are not re-derived here.

What the surplus costs is straightforward: every cubic foot of air you bring in, you heat to flue temperature and throw away, and about four fifths of it is nitrogen that took part in nothing at all.

The quantity that actually tracks the loss

Analysers print an excess air percentage, calculated from measured oxygen as %O2 divided by (20.9 minus %O2), times 100. Use that as a tracking number rather than an absolute, because it is a dry-basis approximation that runs somewhat high on natural gas once the water of combustion has been removed from the sample.

The quantity that is exact, and the one worth reasoning with, is simpler. Call it the flue gas multiple:

Multiple = 20.9 divided by (20.9 minus measured %O2)

For a methane-dominant fuel that multiple is exactly proportional to the volume of dry flue gas produced per unit of fuel burned. At 0% oxygen it is 1.0, meaning a perfect burn. At 7.0% oxygen it is 20.9 divided by 13.9, or 1.504, meaning you are making about half again as much flue gas as a perfect burn would. It is the same number an analyser uses to convert a measured carbon monoxide reading to an air-free basis, which is why it is worth internalising once.

Pair it with net stack temperature, which is measured flue temperature minus the temperature of the air entering the burner, not minus room temperature and not the raw flue reading. Net stack temperature is the temperature rise you actually paid for.

Multiply the two and you have a proportional index of sensible stack loss: how much gas, times how hot you sent it out. It is an index for comparing one setting against another on the same appliance, not an efficiency figure. The analyser's efficiency number additionally accounts for the latent loss in the water of combustion and for fuel-specific constants, which is why it will not match this index and does not need to.

Why a cooler flue can mean a worse appliance

This is the result that surprises people, and it is why stack temperature alone is a trap.

Take one appliance at 7.0% oxygen with a net stack temperature of 320 degrees F. Multiple is 1.504, and the index is 1.504 times 320, or 481.

Now open the air up to 11.0% oxygen. Multiple becomes 20.9 divided by 9.9, or 2.111. All that extra cold air pulls the flue temperature down, so net stack falls to 300 degrees F. The index is 2.111 times 300, or 633.

The flue got 20 degrees F cooler and sensible stack loss went up by 633 divided by 481, or about 32%. A tech watching only a stack thermometer would have recorded an improvement. Both readings are correct. The interpretation is the whole job.

The rule for setting it

State it precisely, because a vague version of this rule is what produces the appliance that fails in January.

Reduce excess air, one step at a time, only while air-free carbon monoxide stays at or below your acceptance threshold across a full firing cycle, at the worst-case building condition, with the appliance at operating temperature.

• Unit of analysis: one appliance, one complete cycle from cold start through steady state, not a spot reading.
• Step size: no more than 1.5 percentage points of measured oxygen per step, with a full cycle re-run after each step. Larger steps hide where the knee is.
• Worst-case condition: every exhaust device in the building running and interior doors in the position that most depressurises the appliance's space. If you only test it in the as-found condition, you have set it for the easy day.
• Acceptance threshold: the listing standards in the ANSI Z21 and CSA series cap carbon monoxide in the products of combustion at 400 ppm air-free for listed gas appliances, so that is the outer limit, not a target. A common shop acceptance is 100 ppm air-free, with any upward trend across the cycle treated as a fault regardless of the peak value. Set your own number and hold it, but do not set it at 400.

Appliance A: the rule says keep going

A power-burner appliance, as-found at 9.6% oxygen, net stack 340 degrees F, air-free carbon monoxide 7 ppm steady.

Index as found: multiple is 20.9 divided by 11.3, or 1.850, times 340 gives 629.

Step one closes the air to 8.2% oxygen. Full cycle re-run: cold-start peak 21 ppm air-free, steady 6 ppm. Under 100, no upward trend, gate passes.

Step two closes to 6.8% oxygen, which is a 1.4 point move. Multiple is 20.9 divided by 14.1, or 1.482. Net stack has fallen slightly to 330 degrees F because the burner is now transferring heat better rather than because it is diluted. Index is 1.482 times 330, or 489. Full cycle: cold-start peak 34 ppm air-free, steady 6 ppm, and repeated with the range hood, dryer and both bath fans running the cold-start peak reaches 52 ppm. All under 100, gate still passes.

Sensible stack loss index went from 629 to 489, which is 489 divided by 629, or a 22% reduction. That is the whole return, and it came from two adjustments and roughly an hour of cycling.

Stop here rather than taking a third step. The remaining margin is small and the insurance you are spending is the appliance's tolerance for a dirty filter, a partially blocked burner or a colder-than-design night.

Appliance B: the rule says stop, and the air was never the problem

A second appliance, same procedure, same rule, opposite conclusion.

As found: 9.4% oxygen, net stack 355 degrees F, air-free carbon monoxide 12 ppm steady. Multiple is 20.9 divided by 11.5, or 1.817, index 645.

Step one to 8.3% oxygen, a 1.1 point move. Multiple 1.659, net stack 350 degrees F, index 581. Air-free carbon monoxide steady at 41 ppm, cold-start peak 96 ppm. That is still under 100, and this is the moment a tech in a hurry books the win.

Step two to 7.2% oxygen. Multiple 1.526, net stack 346 degrees F, index 528. Air-free carbon monoxide steady 148 ppm, cold-start peak 310 ppm. The gate fails here, cleanly and obviously.

But it also failed at step one, and finding that out is the point of the worst-case condition. Re-running step one with the building's exhaust devices on and the mechanical room door closed, the cold-start peak went to 210 ppm. Step one passed only because it was tested on the easy day.

So the appliance goes back to its as-found 9.4% oxygen, and the index that looked like an 18% available reduction, from 645 down to 528, is not available at all. The steepness of that carbon monoxide response to a modest air change is itself the diagnosis: an appliance with sound geometry and a clean exchanger does not go from 12 ppm to 148 ppm over 2.2 points of oxygen. Something is quenching the flame, and until that is found and corrected, no air setting is the answer. Leaning it out would have felt like the same job as appliance A and produced a hazard.

What changes the rule

Condensing appliances. Net stack temperature is low by design and much of the recoverable heat is latent rather than sensible, so the sensible index above understates what is happening and the analyser's own efficiency figure carries more of the story. The carbon monoxide gate is unchanged.

Fixed-air atmospheric appliances. Many natural-draft appliances have no meaningful air adjustment at all. Measured oxygen on those is a diagnostic report about the appliance and the vent, not a setting to chase. Reading a high oxygen figure there and hunting for a damper is a way to break something that was working.

Modulating burners. The rule has to run at more than one firing rate, because the air-fuel relationship is not identical across the range. Test at minimum, mid and maximum fire, and set for the worst of the three rather than the best.

How to verify you got this right

Come back to it cold. Run the full cycle from a genuine cold start with the building in worst case, and record measured oxygen, air-free carbon monoxide, net stack temperature, and the analyser's efficiency figure at the same three points every time: 60 seconds in, at steady state, and just before shutdown. Write those into the job record with the outdoor temperature and the building condition you tested under, because a setting that is correct at 60 degrees F outdoors with the doors open is not verified until someone has seen it at design conditions. If the numbers are not written down with their conditions, the next tech has a setting with no provenance and will either leave it alone or move it blind.

References

• ANSI Z21 and CSA appliance standards, which limit carbon monoxide in the products of combustion to 400 ppm air-free for listed gas appliances
• 29 CFR 1910.1000 Table Z-1, general industry permissible exposure limit for carbon monoxide
• Manufacturer installation and service instructions for permitted burner adjustments, firing rate and sampling port location
• See related: How Combustion Actually Works; How to Read a Flue Gas Result