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.
A thermometer is the one purchase that makes everything else on this site usable. Every temperature on this site, every doneness window, every failure point assumes the reader can take a reading and trust it. So the instrument is worth getting right, and it is also the piece of kitchen equipment most consistently reviewed badly.
The failure mode in thermometer reviews is easy to describe. A manufacturer publishes an accuracy figure and a response-time figure. A review reprints both, adds a paragraph about how the unit feels in the hand, ranks six of them and attaches a purchase link. Nothing was measured. The accuracy figure came from the box.
This page supplies the protocol. A product verdict requires the completed readings, scoring calculation and acquisition disclosure; without them, no ranking or purchase link is justified.
Disclosure, stated before any verdict
This page names no product and gives no verdict. These are the conditions any thermometer verdict should meet, on this site or in any review you read:
- Every unit tested is listed with how it was obtained: purchased, gifted or loaned, stated per unit, in the results table itself rather than in a footer.
- Anything gifted or loaned is labelled on every row where it appears, not once at the top.
- No manufacturer sees results before publication, and any contact with a manufacturer is stated on the page.
- A commission-earning link is disclosed on the page and in our disclosure policy, and it never influences an ordering. Ordering comes from the measured numbers and from nothing else.
- No unit is reviewed by someone who cannot afford to criticise it. If a manufacturer's cooperation was the only way to obtain a unit, the page says so and treats the result with the suspicion it deserves.
A review that cannot meet all of the above has not earned a verdict.
The protocol, in full, before any result
The protocol is fixed and published before the first reading is taken. That matters more than it sounds: a test whose method is decided after the numbers come in can be steered, usually without anyone intending to steer it.
The general house method is on how we test. What follows is the thermometer-specific version.
Sample selection
The sample is six units, chosen across the price range rather than across brands, and chosen before any of them are handled. The selection rule is written down in advance:
| Slot | What it represents |
|---|---|
| 1 | The cheapest instant-read unit widely available in your market |
| 2 | A second cheap unit from a different maker, to separate "cheap" from "that one cheap unit" |
| 3 | A mid-price unit of the kind most often recommended |
| 4 | A second mid-price unit, same reason as slot 2 |
| 5 | A well-known premium unit |
| 6 | A leave-in probe of the type used for long low-temperature cooks |
If the sample changes, the title changes with it. A post called "six thermometers" that tested five is a small lie that makes every other number on the page worth less.
Reference points
Two fixed points are used, both chosen because a home cook can reproduce them without owning a calibration bath.
The ice point. A properly made ice bath sits at the freezing point of water. The bath has to be mostly crushed ice with just enough water to fill the gaps, stirred, and given time to equilibrate before any probe goes in. A glass of cubes with water poured over it is not an ice bath and will read high. The full procedure is the one in how to calibrate a thermometer, and it is run identically for every unit.
The boiling point. Water's boiling temperature depends on pressure and falls as elevation rises. Record location and conditions and use the USDA FSIS boiling-water calibration method with its high-altitude correction guidance, rather than assuming 212°F (100°C).
Both reference points get the same treatment for every unit, in the same session, in the same water, with the units read in a rotating order so that the last unit tested is not always the one whose reading has had the longest time to settle.
Test 1: accuracy at the ice point
Each unit takes a stated number of readings in the ice bath, with the probe positioned per the manufacturer's stated sensor location where one is given, and with immersion depth recorded. Record every reading and the spread, never an average alone. A unit that reads consistently wrong can be corrected by the cook. A unit that reads unpredictably cannot.
Recorded per unit: every individual reading, the deviation from the reference, the spread across readings, and whether the unit offers a user calibration adjustment.
Test 2: accuracy at the boiling point
The same procedure at the other fixed point. This test exists because a unit can be accurate at one end of its range and drift at the other, and because a single-point check is the most common thing a cook does and the least informative thing they could do.
Recorded per unit: the same fields as test 1, plus the deviation between the two fixed points, which is the number that tells you whether the error is an offset or a slope.
Test 3: response time
Response time is where reviews go furthest wrong, because everyone measures it differently and nobody says how. The definition here is fixed in advance and applied identically:
- The probe starts at a stated, stabilised ambient temperature.
- It is inserted into a stirred bath at a stated target, to a stated depth.
- The clock runs from insertion to the moment the display first reaches and holds a reading within a stated tolerance of the bath's true temperature, rather than to the moment the display stops changing.
- The run is repeated a stated number of times per unit, and every run is recorded.
That definition is arguable. Someone else could reasonably define the endpoint differently and get different numbers. What matters is that the definition is stated, so the units can be compared with each other even if the figure cannot be compared with someone else's marketing figure.
Two things get recorded that reviews almost never mention: whether the display resolution is fine enough to see the approach at all, and whether the unit smooths its display in software, which makes a slow sensor look fast.
Test 4: drift after a period of real use
Every unit goes into normal kitchen use for a stated period, in a stated set of conditions, then returns to both fixed points and repeats tests 1 and 2 unchanged. The comparison is against that unit's own opening numbers.
This is the test that separates an instrument from a gadget, and it is the reason the full protocol takes months rather than an afternoon. A unit that is accurate out of the box and drifts is worse than a unit that is slightly off and stays put, because the second one can be trusted after a calibration check and the first one cannot be trusted at all.
How each unit is obtained
State it per unit in the results table: purchased, gifted or loaned.
How the verdict is reached
The decision rule is written down here, before the numbers exist, so that it cannot be reverse-engineered to fit whichever unit turns out to be likeable:
- Accuracy at both fixed points is the primary criterion, weighted by consistency rather than by best-case reading.
- Response time is secondary, and only matters above a threshold of practical usefulness. Beyond a certain speed the difference stops changing what a cook can do.
- Drift is a disqualifier, not a ranking factor. A unit that will not hold its calibration does not place, whatever else it does.
- Price is not a criterion. It appears in the verdict only in the form of the question "does the extra money buy a measurable difference in 1, 2 or 3."
Nothing is ranked by build quality, brand reputation, packaging, app features or how it feels in the hand. Those are real considerations and they belong to the reader, not to a measurement post.
Reader worksheet
Each section below lists what to record.
Accuracy at the ice point
Record, per unit: obtained (purchased / gifted / loaned), every individual reading, deviation from the corrected reference, spread, user-calibration available yes/no.
Accuracy at the boiling point
Record the same fields, plus the offset-versus-slope comparison between the two fixed points.
Response time
Record, per unit: every timed run, the median, the worst run, display resolution, and whether display smoothing was observed.
Drift after a period of real use
Record, per unit: opening deviation at both fixed points, closing deviation at both, the change, and the conditions the unit lived in during the period.
The verdict
A verdict from those tables answers three questions, in this order.
Which units are accurate enough to trust for the cooking this site covers? Possibly all of them. Possibly none. Gentle cooking is where instrument error hurts most, because the interesting decisions happen inside windows only a few degrees wide, and an error that is invisible when you are checking a roast is decisive when you are holding a custard.
Is the cheapest unit good enough? A real question with a real chance of a yes. If the inexpensive unit is accurate at both points, holds its calibration and responds fast enough to be usable, the extra money is buying convenience and durability, and this protocol does not measure those. Judge the unit against those recorded checks instead of assuming price determines accuracy.
What does the extra money buy, if anything? Stated as a measured difference or not stated at all.
Negative findings get the same space and the same position on the page as positive ones. If a widely recommended unit fails the drift test, that finding leads the section.
What this protocol does not test
This list is as important as the results, and it is deliberately long.
- Durability. We do not drop, crush, or stress-test any unit, and a testing period is not a lifetime. A unit that survives the test may fail in a year.
- Waterproofing. Manufacturers publish ingress ratings. We do not verify them, we do not submerge units to check, and we do not repeat the claim as though we had. If waterproofing matters to you, the manufacturer's rating is the only information available and it is not ours.
- Long-term reliability. Battery life, connector wear, screen failure, hinge failure on folding probes, and the behaviour of the electronics after a few hundred cycles are all outside a bench test of this length.
- Behaviour at extremes. The protocol covers the range this site cooks in. Very high-temperature work and freezer-range work are not covered and the results should not be extended into them.
- Wireless features, apps and alarms. Not measured, not ranked, not commented on.
- Sample-to-sample variation. The protocol tests one unit per model. Manufacturing variation between two units of the same model is real and the test says nothing about it. This is the single largest limitation of the whole exercise, and it applies to every review of this kind including the ones that do not mention it.
- Customer service, warranty handling and returns. Not tested, not our field.
FAQ
If a unit reads consistently wrong, is it useless? Not necessarily. A consistent offset can be corrected, either through the unit's own calibration adjustment or by writing the offset on a piece of tape and doing the arithmetic. What cannot be corrected is inconsistency. That is why the protocol records the spread across readings rather than only the average.
Why fixed points instead of comparing against a reference thermometer? Because a reference thermometer needs its own traceable calibration, which most kitchens do not have, and because the ice point and the boiling point are reproducible in any kitchen. Anyone can repeat the fixed-point accuracy test. They cannot repeat a comparison against an instrument they do not own.
Does the protocol name a winner? Only if the numbers name one. Treat units that are indistinguishable inside the precision of the method as indistinguishable rather than manufacturing an ordering from noise.
In Plain English
This post is a measurement protocol, not a product review. It explains how to compare thermometers at two fixed points, measure response time, and re-check accuracy after use. Apply the stated scoring rule only to units you actually tested.
It helps you if you are about to buy a thermometer, or if you own one and have never checked it. The moment to use it is before you spend anything, because the most likely useful outcome of this test is that the cheap unit is good enough and the money is better spent elsewhere.
What to do next: check the thermometer you already own. That takes about two minutes, costs nothing, and tells you more about your cooking than any purchase would. How to calibrate a thermometer has the procedure. If you own no thermometer at all, what to buy first covers the order to acquire things in, and it does not start with the expensive items.
Before you buy anything
Check the thermometer you already own first; the protocol above is for comparing units, not a reason to buy one.
Before applying the protocol, read our disclosure policy. It states how equipment is obtained, what happens when something is gifted, which pages take no commercial relationship of any kind, and how an inconvenient result is handled. This page makes no product verdict.
Related technique: Twelve Ingredients Worth Paying More For, And Eight That Aren't.