How to calibrate a thermometer: ice point, boiling point and comparison calibration

To calibrate a thermometer at home or in a kitchen, use the ice-point method: pack a glass with crushed ice, add water to the top and stir, put the probe in the middle at least 2 inches deep without touching the sides or bottom, wait at least 30 seconds, and check that it reads 32 °F (0 °C) (USDA FSIS). A food thermometer passes if it reads within ±2 °F, or ±1 °C if it reads in Celsius (FDA Food Code, 4-203.11). If it is off and has an adjustment nut or calibration button, adjust it while it is still in the ice. That check tests one temperature. A full calibration compares the thermometer with a calibrated reference at several temperatures across the range you use, and records the error and its uncertainty.

This page covers both: the ice and boiling-water checks done properly, what to do when a thermometer reads wrong, and then the comparison procedure a lab or in-house calibration room follows, with a worked example, the records to keep and how often to recalibrate.

How to calibrate a thermometer with ice water (ice-point method)

The ice point is the most reliable check you can make without special equipment, because a properly made mixture of ice and water sits at 0 °C without needing another thermometer to tell you so. NIST recommends it as a secondary fixed point for calibrating thermometers, reproducible in a lab to about 0.002 °C when done carefully (NIST SP 1088, §7.2). The catch is “properly made”.

1. Fill a large glass with finely crushed ice. USDA specifies finely crushed ice, and NIST shaves it to particles of 2 to 5 mm for lab work (NIST SP 1088, §7.7): small pieces pack tightly, with no pockets of plain water.
2. Add clean tap water to the top of the ice and stir well. The ice point is ice saturated with water: the ice should fill the glass, not float in a layer of water (NIST SP 1088, §7.7).
3. Insert the stem or probe at least 2 inches, in the middle, touching neither the sides nor the bottom of the glass. A dial (bimetal) thermometer senses along the last 2 to 2½ inches of the stem, so the whole sensing area has to be in the slush (USDA FSIS).
4. Wait at least 30 seconds (longer if the reading is still changing).
5. Read it. It should show 32 °F (0 °C).
6. Adjust if you can and need to. On most dial thermometers, hold the adjusting nut under the dial with a suitable tool and turn the head until the pointer reads 32 °F, without taking the stem out of the ice (USDA FSIS). Digital models with a calibration or reset function have their own steps: follow the manual.

What counts as a pass? The FDA Food Code, the model code FDA offers to state and local regulators for restaurants and retail food, requires food thermometers to be accurate to ±1 °C if they read in Celsius or in both scales, and ±2 °F if they read only in Fahrenheit, over their intended range of use (4-203.11). Thermometers for ambient air and water, such as the one in a refrigerator, may be off by up to ±1.5 °C or ±3 °F (4-203.12). The 2026 edition, published in September 2026, keeps the limits of the 2022 edition. Your state or local code decides which edition applies.

The lab version of the ice point

For checking reference and liquid-in-glass thermometers, NIST describes a stricter recipe (NIST SP 1088, §7.7–7.8):

• Ice made from distilled or deionized water (or the clear part of commercial ice), shaved to particles of 2 to 5 mm, packed in a vacuum (Dewar) flask and saturated with distilled water.
• After 15 to 30 minutes, siphon off the meltwater and add ice. The bath is ready when the ice is saturated but not floating and there is no water on top. Make a fresh one every day.
• Clean the thermometer and your hands (salt from skin contaminates the bath), pre-chill the thermometer, and leave at least 3 cm of slush below the bulb.
• Read a mercury thermometer after about three minutes, an organic-liquid one after about 15 minutes, looking perpendicular to the stem to avoid parallax.

Ice made from tap water is often good enough for routine work: EURAMET notes that in many countries it gives an ice point between −0.025 °C and 0 °C, although deionized water is best (EURAMET cg-8, §4.1).

Boiling-water check: correct for altitude and pressure

The boiling-water method works the same way as the ice point: immerse the stem at least 2 inches in water at a full rolling boil, wait at least 30 seconds and compare with the boiling point (USDA FSIS). The problem is the boiling point itself. Water boils at 212 °F only in distilled water at a pressure of one standard atmosphere, 29.921 inHg; USDA notes that tap water under unknown conditions will probably boil at least 2 °F, and perhaps as much as 5 °F, lower (USDA FSIS). Strictly, pure water at standard pressure boils at 99.974 °C (211.95 °F) on today’s temperature scale (NIST Chemistry WebBook), which rounds to 212 °F.

Pressure falls with altitude, and the boiling point falls with it:

That is roughly 1.8 °F less for every 1,000 ft, in line with USDA’s rule of thumb of just under 1 °F for every 500 ft (USDA FSIS, High Altitude Cooking). Two more things move the result:

• The weather. The table assumes the standard atmosphere, but the actual pressure changes from day to day. Near 100 °C, the boiling point changes by about 0.28 °C (0.5 °F) for every kPa (0.3 inHg) of pressure (NIST Chemistry WebBook). If you know the actual pressure, the table’s pressure column gives the boiling point more accurately than the altitude does.
• Which pressure. The pressure in weather reports is usually corrected to sea level (the altimeter setting). Boiling depends on the actual pressure where you are, the station pressure, which is lower at altitude (NOAA NWS, Station Pressure).

Use the boiling-water check for thermometers you use at high temperatures (candy, frying), and treat it as a rough one. The ice point has none of these problems.

What to do if your thermometer reads wrong

First rule out the method: a bath with too much water, a probe touching the glass, or a reading taken too early. Then you have three options (USDA FSIS):

• Adjust it, if it has an adjustment: a calibration nut on dial thermometers, a calibration or reset function on some digital ones, or, on some liquid-filled thermometers, moving the glass stem in its holder.
• Allow for the error. USDA’s example: a thermometer that reads 214 °F in boiling water reads 2 °F high, so you subtract 2 °F from its readings, or cook ground beef to 162 °F on it instead of 160 °F. This only works if the error is the same at the temperatures you use, which a one-point check can’t prove.
• Replace it.

Not every thermometer can be adjusted, and each type has its own weak points:

For a digital thermometer with interchangeable probes, remember that the probe and the display are two parts. A calibrator that simulates the sensor’s electrical signal only tests the display, not the probe (EURAMET cg-11, §1.2). Check the probe and display together, in a bath or block.

Before blaming the thermometer, also check how you use it: the center of a food is usually cooler than the surface, so the tip belongs in the center of the thickest part (USDA FSIS), and a Food Code establishment also needs a thin-probe thermometer for thin foods such as burger patties and fish fillets (4-302.12).

Thermometer calibration vs a check

Calibration and adjustment are different operations: calibration compares the thermometer with the reference and records the error; adjustment changes the thermometer, and after an adjustment it usually has to be calibrated again (VIM 2.39, 3.11). Our guide to what calibration is explains the difference.

“NIST traceable” on the box is a claim that the calibration links to NIST standards through an unbroken chain of calibrations with stated uncertainties. NIST does not certify other organizations’ traceability claims: the supplier has to support the claim, and the user has to judge it (NIST traceability policy). What you want to see is a certificate with results at stated temperatures and their uncertainty.

What you need to calibrate a thermometer

• Reference thermometer

  A platinum resistance thermometer (PRT, or SPRT for the lowest uncertainty) with a calibrated readout, or a calibrated reference thermometer

  Calibrated by an accredited lab. A lab should have two reference standards per range, or one plus at least two fixed points such as the ice point, to catch drift.

  DKD-R 5-1, §4 and §9

• Accuracy ratio

  Reference system uncertainty a small fraction of the tolerance being checked

  A common target is 4:1: for a ±1 °C food thermometer, about ±0.25 °C or better. Check yours with the TUR calculator.

• Temperature source

  A stirred liquid bath or a dry block calibrator, characterized for uniformity and stability

  Both thermometers must sit in the useful volume, where the temperature is uniform and stable. In a dry block, the hole or insert should be at most 0.5 mm wider than the probe (up to 660 °C).

  DKD-R 5-1, §7; EURAMET cg-13, §2.2

• Ice point

  Crushed ice and water in an insulated flask

  For the as-found check of the thermometer and as an intermediate check of the reference.

  NIST SP 1088, §7.7; DKD-R 5-1, §9

• Environment

  Ambient conditions within the specifications of the readout and electrical equipment

  Monitor and record them if they affect the result.

  DKD-R 5-1, §4

• Paperwork

  The thermometer specification (range, tolerance), your procedure and a record form

  Know the tolerance and the range of use before you start. See how to write a calibration procedure.

The temperature source matters as much as the reference. The calibration of a dry block mainly relates to the temperature of the block: the thermometer you put in it can be at a different temperature, especially with a different insert or probe than in the block’s calibration (EURAMET cg-13, Annex C). That is why a comparison calibration reads an external reference thermometer in the same block, as in DKD’s worked example, rather than the block’s own display (DKD-R 5-1, Appendix A.3).

Test points and number of readings

DKD’s guideline for resistance thermometers asks for at least two temperatures, with the points and range agreed with the customer, and states that results are valid only between the lowest and highest calibrated temperature (DKD-R 5-1, §1 and §5). The Food Code’s accuracy requirement applies across the whole range of use, from refrigeration to cooking temperatures if the thermometer is used for both (Food Code, Annex 3, 4-203.11). Put together:

• Cover the lowest and highest temperatures you measure, plus at least one point in between.
• Include 0 °C when it is in the range: it is the easiest point to realize and the one you will use for checks between calibrations. EURAMET asks for it with thermocouples (EURAMET cg-8, §12.6).
• Calibrate thermocouples from the highest temperature down (EURAMET cg-8, §10.1). For industrial platinum thermometers, rising and falling temperatures can give different results (hysteresis), so a lab tests both directions at mid-range when it needs that figure (DKD-R 5-1, §8.10).

The plan for the example thermometer below, a probe used from freezer to cooking temperatures:

How many readings? No free standard fixes a number for general thermometers. The worked example in DKD-R 5-1 averages 20 readings taken every 30 seconds over 10 minutes after an hour of stabilization (Appendix A.3). Whatever you choose, write it into your procedure and use the spread of the readings in the uncertainty.

Step-by-step thermometer calibration procedure

The sequence follows DKD-R 5-1 and EURAMET cg-13, plus the as-found and as-left logic of any calibration. Adjust it to your thermometer type and write your choices into your calibration procedure.

1. 1

   Identify and inspect

   Record ID, model, range, resolution and tolerance. Check the probe, cable and display; for glass thermometers, look for separated columns, bubbles and cracks with a magnifier.

   Damaged: tag out, record as not calibratable

2. 2

   Ice-point check

   Read the thermometer in an ice point. For reference and resistance thermometers, a change since the last calibration is the first sign of damage from shock or transport (DKD-R 5-1, §5–6). Wait 72 hours after use at high temperature for glass thermometers (NIST SP 1088, §7.4).

3. 3

   Mount

   Place the thermometer and the reference side by side in the bath or block, sensing elements at the same depth in the useful volume. As a first estimate, immerse at least 10 times the probe diameter, measured from the top of the sensing element: that keeps the error below about 1 % of the difference between the bath and room temperature, and each further 10 diameters cuts it by a factor of 10 (DKD-R 5-1, §8.3).

4. 4

   Stabilize

   Set the first temperature and wait until both readings are stable. Take readings of both thermometers at the same time, or average them over the same period (DKD-R 5-1, §7).

5. 5

   As-found readings

   At each test point, record the reference temperature and the reading of the thermometer under test.

6. 6

   Check immersion

   Withdraw the thermometer slightly, about 10 % of the depth or 20 mm in a dry block. DKD's criterion: the change should not exceed 10 % of the uncertainty you are aiming for; if it does, heat conduction along the stem is affecting the reading and its uncertainty contribution must be increased (DKD-R 5-1, §8.3; EURAMET cg-13, Annex C).

7. 7

   Evaluate

   Error = reading − reference temperature, at every point. Compare it with the tolerance.

   Outside tolerance: record as found out of tolerance, open an out-of-tolerance report

8. 8

   Adjust only if needed

   Use the manufacturer's adjustment method, and only past the threshold your procedure sets.

9. 9

   As-left readings

   After any adjustment, repeat all points as a separate set. Report both: DKD requires results before and after adjustment, especially for direct-reading thermometers (DKD-R 5-1, §10).

10. 10

    Uncertainty, record and label

    Evaluate the uncertainty at each point, issue the certificate or record, and label the thermometer.

A contact thermometer only measures its own temperature. The calibration tells you how its reading relates to the temperature of its sensor; making sure the sensor is at the temperature you want to measure, in use, is up to the user (DKD-R 5-1, §2).

Acceptance criteria and measurement uncertainty

Worked example: passes at the ice point, fails at 100 °C

The example thermometer, TH-117, is a handheld digital thermometer with a Pt100 probe, −50 to 150 °C with 0.1 °C resolution, used in a food plant. Its tolerance is the Food Code limit, ±1 °C over the range it is used in. It was compared with a reference PRT in a stirred bath at four points, three readings each. The error of each reading is the indication minus the reference temperature. The values are invented for this example.

How to read it:

• As found, 0 °C: +0.1 to +0.2 °C. An ice-point check would have passed this thermometer.
• As found, 50 and 100 °C: the error grows with temperature, to +0.6 °C at 50 °C and +1.2 to +1.3 °C at 100 °C, outside ±1 °C. This is a span (slope) error, the kind a one-point check can’t see. The out-of-tolerance report has to ask what was measured above about 80 °C with this thermometer since its last good calibration.
• As left: after a two-point adjustment in the instrument’s calibration menu, every reading is within ±0.3 °C.

Close to the limit, whether a result counts as a pass depends on the uncertainty and on your decision rule (Global ACI-TECH-1-002, formerly ILAC G8). Our guard banding guide explains the options.

What goes into the uncertainty

The uncertainty of the calibration combines the reference and the source with contributions from the thermometer itself, which can be many times larger than the reference’s (DKD-R 5-1, §8). DKD’s worked example for a digital thermometer with a Pt100 probe, calibrated against a working-standard PRT in a dry block at about 350 °C, gives a real budget (DKD-R 5-1, Appendix A.3):

The thermometer read 0.23 K high, with an expanded uncertainty of 0.25 K. Hysteresis, a property of the probe, was the largest single contribution; DKD notes it often is (DKD-R 5-1, §1). Build or check your own budget with the uncertainty calculator.

What to record on a thermometer calibration certificate

Whether you issue a certificate or fill in an internal record, keep:

• Identification: thermometer and probe IDs or serial numbers, manufacturer, model, range, resolution.
• Method: your procedure number and revision, and the guideline it follows (for example, DKD-R 5-1).
• Equipment: the reference thermometer and readout with their traceability, the bath or block, and the insert used.
• Conditions: immersion depth (DKD asks for it on the certificate so the calibration can be repeated, §8.3), ambient conditions, stabilization time.
• Results per point: reference temperature, readings or their mean, error, as found and as left.
• Uncertainty: the expanded uncertainty at each point, and for resistance thermometers the self-heating and hysteresis behavior and the measuring current (DKD-R 5-1, §10).
• Decision: pass or fail against the stated tolerance, the decision rule used, and any adjustment made.
• Who and when: technician, date, next due date.

Our guides to calibration certificates and calibration records cover each field in detail. Print a matching status sticker with the free calibration label generator.

How often to calibrate a thermometer

There is no universal interval, but the sources agree on the logic:

• Food thermometers: calibrate according to the manufacturer’s specifications as often as necessary to keep them accurate (Food Code, 4-502.11(B)). In practice, that means a regular ice-point check and a check after any drop or shock.
• Liquid-in-glass thermometers: check at the ice point at least once a year even when used near room temperature. New ones, or ones used at high temperatures, may start with checks as often as monthly, then every two months, then once or twice a year if the ice point stays stable (NIST SP 1088, §4 and §7.6). Mercury thermometers used at 200 °C or below can be recalibrated at the ice point alone; above 200 °C they need full recalibration, and NIST does not recommend single-point recalibration for organic-liquid thermometers (NIST SP 1088, §1).
• Reference thermometers: DKD recommends recalibration at most every 2 to 3 years for SPRTs and every 1 to 2 years for industrial PRTs, depending on their history, with ice-point checks in between (DKD-R 5-1, §9, Table 1).
• Thermocouples: there is no formal frequency; EURAMET recommends a checking and recalibration program built on your own experience, with in-place checks against a reference for installed thermocouples (EURAMET cg-8, §13).

Then let your own data set the interval: extend it if thermometers keep coming back in tolerance, shorten it if they don’t. Our calibration intervals guide explains the methods.

When a thermometer fails calibration

An as-found result outside tolerance means every measurement made with the thermometer since its last good calibration may be wrong, at the temperatures where it failed. In the example, the thermometer read more than 1 °C high near 100 °C: food it showed at a safe cooking temperature could have been more than 1 °C cooler. Our guide to out-of-tolerance results walks through the impact assessment.

How to calibrate an infrared thermometer

Infrared thermometers measure the thermal radiation from a surface, so they are calibrated against a radiation source of known temperature, a blackbody or flat-plate calibrator, not in a bath. ASTM E2847 is the US test method for handheld, direct-reading, wideband infrared thermometers that measure below 1000 °C, including the uncertainty calculation and the report (ASTM E2847-21(2026), scope). In practice (Fluke Calibration):

• Emissivity: set the thermometer’s emissivity to that of the source, or correct for the difference.
• Distance and alignment: the thermometer’s field of view must fall inside the source; aim it square to the surface, within 5°.
• Points: one point if the thermometer is used over a narrow range, three (minimum, middle and maximum) for a wide range.
• Reflected heat: keep other heat sources and people away from the target, especially at low temperatures.

Fluke lists emissivity as a major source of uncertainty in the calibration; in use, too, the reading depends on the emissivity setting matching the surface you measure.

Common mistakes when checking a thermometer

Related guides

• How to write a calibration procedure: the general structure this procedure follows.
• Thermometer calibration hub: everything on thermometers in one place.
• FDA calibration requirements: what FDA rules ask of calibration programs.
• TUR calculator: check that your reference is good enough for the thermometer.
• Uncertainty calculator: build the uncertainty budget.
• Out of tolerance: what to do when a calibration fails.
• Calibration intervals: how often to recalibrate.
• Torque wrench calibration: the same structure applied to torque tools.