Scope. This page provides a controlled reader-run protocol and does not claim site-run measurements. For cooking, cooling, storage and reheating, follow the current official guidance linked on the page.
Every number on this site depends on somebody having done this first. So does every number you take in your own kitchen. A probe is an instrument, instruments drift, and a reading from an instrument nobody has checked is not a measurement. It is a suggestion with a decimal place.
The check takes two tests. Together they take a couple of minutes, most of which is waiting. You need a glass, some ice, some water, a pan and the probe.
Why an unchecked probe makes every other number meaningless
Consider what a small offset does across the four bands.
An offset of a couple of degrees is invisible when you are boiling pasta. In the 60–70°C band, where fish and eggs and tender cuts live, the entire doneness window can be narrower than the error you are carrying. You pull the fillet at what your probe calls the target, and you are actually somewhere else, consistently, every time, in the same direction. The result is worse than random error, because a consistent bias trains you to distrust the correct instruction rather than the faulty instrument. You conclude the recipe is wrong. You adjust. You get further from the truth.
There is a second, less obvious problem. Probes rarely have a single constant offset across their whole range. A probe can be close to correct at one end of the scale and further out at the other. That is why the routine below uses two reference points rather than one: two points tell you both how far off it is and whether the error changes with temperature.
The ice-point check
The ice point is the more useful of the two checks, for two reasons. It is the easier one to get right, and it sits far closer to the low-temperature work this site is about.
It also has one very common failure mode, which is the whole reason this section exists.
A correct ice bath is mostly ice with just enough water to fill the gaps. Not a glass of cold water with some ice floating in it. The mixture has to be a genuine slurry, packed, so that the water is in continuous contact with melting ice throughout. A glass of water with a few cubes bobbing on top will read warm, will drift as you watch it, and will give you an offset that is entirely fictional. Cooks who run this test and get a puzzling result have usually made this mistake.
The procedure:
- Fill a tall glass or a small jug with crushed or small ice, right to the top, and pack it down. Crushed is better than cubes because it removes air pockets.
- Add cold water until it just reaches the top of the ice, no further. Stir. Top up with more ice if the level drops.
- Let it stand for a minute or two. The slurry needs to equilibrate; the reading is not valid on a bath you have just built.
- Insert the probe into the middle of the slurry, not touching the sides or the bottom of the glass. Glass walls conduct room heat.
- Stir gently with the probe and wait for the reading to settle. Settling time varies enormously between instruments, and how long it takes is itself worth noting.
- Read it. A correctly built ice bath sits at the freezing point of water, which is the zero of the Celsius scale by definition, and 32°F. This is a property of the scale rather than one of our measurements, which is why it is the reference everyone uses.
- Write down the difference between the reading and zero, with its sign. That is your low-end offset.
Repeat the check two or three times, rebuilding the bath each time. If the offset moves between attempts, your bath construction is the likeliest culprit; a probe with genuinely unstable behaviour at the ice point is a probe to replace.
The boiling-point check, and the altitude correction
The second reference point is water at a rolling boil. This one is more useful for the upper bands and less trustworthy overall, because it depends on something in your environment that you do not control.
Water boils at the temperature at which its vapour pressure equals the surrounding air pressure. Air pressure falls with altitude and moves with the weather. So the boiling point in your kitchen is not a fixed number, and anyone who prints a single boiling-point figure without stating a pressure or an altitude is printing a number that is only true near sea level on an ordinary day.
USDA FSIS explains that boiling temperature falls as elevation rises and gives representative values in its high-altitude cooking guidance. For a boiling-point check, use that authority or a local Extension/health-department value appropriate to your elevation rather than assuming 212°F (100°C).
The procedure:
- Bring a deep pan of water to a full, sustained rolling boil. Deep matters, because you need to get the probe tip into the body of the water rather than near the base, where a hot spot will read high.
- Hold the probe in the water, away from the base and the sides, and let it settle.
- Compare the reading to the boiling point for your altitude, following the USDA FSIS boiling-water calibration method and its linked altitude caveat.
- Write down the difference, with its sign. That is your high-end offset.
Two practical notes. Steam is hot, and a probe held over a rolling boil for a minute is a probe held in a steam plume, so mind your hand. And if you do not know your altitude with any confidence, treat this check as a rough sanity test rather than a calibration, and rely on the ice point for the number.
Comparing the two offsets
Once you have both numbers, you know more than most people do about their own kitchen.
| What you find | What it means | What to do |
|---|---|---|
| Both offsets near zero | The instrument is behaving | Re-check periodically and get on with cooking |
| Both offsets similar and non-zero | A constant bias across the range | Adjust if the probe allows it; otherwise compensate by the same amount everywhere |
| Offsets differ noticeably from each other | The error changes with temperature | Compensate using the offset from the end of the range you actually work in, and treat the other end with suspicion |
| Readings will not settle, or move between attempts | The instrument is unreliable rather than merely inaccurate | Replace it; a probe you cannot trust to repeat is not compensable |
That last row is the one worth internalising. Inaccuracy you can correct. Imprecision you cannot, because there is no fixed number to correct by.
Adjustable, or only compensable
There are two kinds of probe and the difference decides what you do next.
Adjustable probes can be reset so that the ice-point reading is corrected in the instrument itself. On analogue dial thermometers this is usually a nut under the dial head that you hold with pliers while turning the dial to zero, with the stem still in the ice bath. On some digital probes there is a calibration function in the menu, often requiring the probe to be held in an ice bath while it is invoked. Consult the instrument's own instructions rather than ours.
Non-adjustable probes simply read what they read. These are not useless. A probe with a known, stable offset is a perfectly good instrument as long as you apply the correction every single time, without exception, including the time you are in a hurry.
Which is precisely why the offset has to live somewhere permanent rather than in your head.
Write the offset down somewhere permanent
Put it on the probe. A strip of tape on the handle with the offset and the date you measured it is the single highest-return thirty seconds in this whole article. Do it now, while the numbers are in front of you.
Do it because the alternative is remembering, and remembering fails in exactly the situation where the number matters: mid-cook, distracted, with a probe you last checked four months ago and an expensive piece of fish approaching the top of its window.
If you own more than one probe, label all of them, and note which is which. Two probes disagreeing with each other is common. Two probes disagreeing while both are labelled with their own verified offset is a solvable problem.
How often to re-check, and what drift looks like
Re-check on a schedule rather than on suspicion, because drift is gradual and you will not notice it happening.
Sensible triggers for a re-check:
- After the probe is dropped, which is the commonest cause of a sudden change.
- After it goes in the dishwasher, or gets submerged when it should not have been.
- After a battery change on a digital instrument.
- After long storage.
- Periodically regardless, at whatever interval you will actually keep to. A schedule you follow beats a better schedule you ignore.
Drift shows up as a slow migration of the ice-point reading in one direction over successive checks. A probe that read close to zero last year and is now consistently off in the same direction is drifting, and the useful response is to keep updating the tape on the handle. When the offset gets large enough that you no longer trust the compensation, the instrument has told you it is finished.
What a typical offset looks like
Use this protocol rather than a claimed "typical" error. This page supplies no representative instrument figure, because an anecdotal sample would be exactly the kind of unsourced number the method is designed to replace.
The companion six-thermometer protocol applies these same two checks and defines fields for accuracy at both reference points, response time, and drift across the test period. It names no product and gives no verdict. The honest answer to "how far out is my probe likely to be" is that yours is the only sample that matters, and you now know how to check it.
In Plain English
Calibration means checking your thermometer against two known reference points, the ice point and the boiling point, and recording how far off it reads.
It matters for anybody who cooks by number rather than by eye, and it matters most in the narrow bands where fish, eggs and custards live, because there the instrument's error can be as wide as the entire window you are aiming at.
Do it when you buy a probe, after you drop it, after a battery change, and on a regular schedule after that. The whole job is: build a proper ice slurry, read it, note the difference from zero; boil a deep pan, read it, compare against the cited boiling point for your altitude; then either adjust the instrument or write the offset on a piece of tape on the handle and apply it every time.
What to do next: go and check the probe in your drawer. If you have never done it, the result will tell you whether every number you have taken so far meant what you thought it meant.
Where to go from here
Print the one-page calibration guide and keep it in the drawer with the probe. It carries both checks, the ice-slurry warning, a space for your two offsets and the date, and nothing else. It is free, ungated, and sits with the rest of the free printables.
The log format, and the rule for a probe that drifts mid-session, are set out in how we test.