How to calibrate a scale: test weights, eccentricity, linearity and repeatability

To calibrate a scale, you place weights of known, traceable mass on it under controlled conditions and record how far its readings are from those masses: at several loads across its range, repeatedly at one load, and at different positions on the pan. Then you work out the uncertainty of those results and decide whether the scale is good enough for its use. That is the method in EURAMET cg-18, the free guide that Europe’s national metrology institutes wrote for calibration laboratories (cg-18, §4.1 and §5). Pressing the CAL button on a kitchen or lab scale with one weight is something else: an adjustment that resets the scale at one point. It is useful, but on its own it tells you nothing about the rest of the range.

This page covers both: what you can do at home with a digital scale and what that doesn’t prove, then the full procedure for scales and balances in quality, lab and pharmaceutical work, with two worked examples, USP <41> and how calibration differs from legal-for-trade verification.

Can you calibrate a scale at home?

You can check it, and you can adjust it if it has a calibration mode. Whether the result counts as a calibration depends on what you compare it with and what you record:

How to use the CAL button on a digital scale

The exact sequence depends on the model, so the manual comes first. The pattern is usually the same:

1. Set the scale up properly. A hard, level, stable surface, away from drafts, vents and vibration. Level the scale if it has a bubble.
2. Warm it up. Leave it switched on for the time the manual gives. One lab balance guide, for example, asks for five minutes after the balance has reached room temperature (OHAUS, Navigator Calibration Guide).
3. Enter calibration mode with the button or key combination in the manual. The scale zeros itself with the pan empty.
4. Place the mass it asks for in the center of the pan when prompted. That is often the capacity or a round value; a “span” calibration uses zero plus this one mass, and a “linearity” calibration uses zero, mid-scale and full scale (OHAUS).
5. Wait for the confirmation, remove the weight and check the scale with a different known weight if you have one. Checking with the same weight you used to adjust it proves nothing about that weight.

The adjustment can only be as good as the weight. A “100 g” weight bought online without a certificate may be fine, or it may be off by more than the scale’s resolution; you don’t know until it is compared with a calibrated one.

What you can test without calibration weights

Two of the three standard tests don’t need a known mass. EURAMET cg-18 allows loads that are not calibrated for the repeatability and eccentricity tests, as long as their mass stays constant during the test (cg-18, §4.3, §5.1 and §5.3). So with any stable object (a full can, a stack of coins taped together):

• Repeatability: put it on and take it off ten times, zeroing in between, and note the spread. Compare it with the repeatability in the scale’s specification, if it has one.
• Off-center loading: put it in the center, then at the front left, back left, back right and front right of the pan. The readings should be the same.

What you can’t check without a known mass is whether the number is right. Coins and food packages have nominal masses, not measured ones, so they can tell you that a scale is grossly wrong, not that it is accurate.

Calibration, adjustment and legal-for-trade verification

Three operations get called “calibration”. They are different, and confusing them is behind many audit findings and arguments with service companies.

• Calibration compares the scale with traceable test weights and states the errors and their uncertainty. It does not change the scale (VIM 2.39). EURAMET cg-18 describes it as applying test loads under specified conditions, determining the error or variation of the indication and evaluating the uncertainty (cg-18, §4.1).
• Adjustment changes the scale so it reads closer to the reference: the CAL routine, a built-in (“internal calibration”) weight, or a technician’s span and linearity settings. After an adjustment the scale usually has to be calibrated again (VIM 3.11, Note 3). A scale that is normally adjusted before use should be adjusted before the calibration too, with the means the user normally applies, unless agreed otherwise; for high-resolution balances, cg-18 recommends adjusting immediately before the calibration and before use (cg-18, §4.1.3).
• Legal verification is the official check that a commercial scale (one used to sell, buy or charge by weight) meets the legal tolerances. In the US the rules are in NIST Handbook 44, adopted by all 50 states, and a commercial scale needs an active NTEP Certificate of Conformance plus approval by the state where it is used (NIST OWM, Weighing FAQs). Internationally, OIML R 76 sets the requirements for non-automatic weighing instruments.

Internal vs external calibration. A balance with a built-in weight adjusts itself, sometimes automatically when it is switched on or when the temperature changes (EURAMET’s example balance does both; cg-18, App. H1.1). That keeps the sensitivity right between calibrations, but it is a one-point adjustment with a weight inside the balance: it doesn’t check the other loads, repeatability or off-center errors. The calibration certificate should state whether the balance was adjusted internally or externally, and with which weight (cg-18, §8.2).

Legal tolerance vs calibration. A calibration has no built-in pass/fail limit: it reports errors. Pass or fail comes from a tolerance you apply, which can be your process requirement, the manufacturer’s specification, or the legal tolerance for the scale’s accuracy class. Handbook 44 and OIML R 76 express that tolerance in verification scale intervals (e) and let it step up with the load:

Maintenance tolerances apply to scales in use; acceptance tolerances, half as large, apply to scales put into commercial use for the first time, and to the first official test within 30 days of being placed in service or of corrective service, major reconditioning or overhaul (Handbook 44, G-T.1, G-T.2 and 2.20 T.N.3.2). OIML R 76 works the other way round: its table gives 0.5 e, 1 e and 1.5 e for initial verification, and the limits in service are twice those (R 76, 3.5.1, 3.5.2).

What you need to calibrate a scale

• Test weights

  Standard weights with a valid calibration certificate, traceable to the SI

  Either a measured conventional mass with its uncertainty, or confirmation that each weight is within its maximum permissible error. Weights that meet OIML R 111 also meet the density, surface and magnetism requirements.

  EURAMET cg-18, §4.3.1

• Accuracy ratio

  Weight error (plus uncertainty, if used at nominal value) below one-third of the tolerance at that load

  The rule in Handbook 44 (App. A 3.2) and OIML R 76 (3.7.1), and in USP <41> as written until January 2026. Check your ratio with the TUR calculator.

  Handbook 44, App. A 3.2; OIML R 76-1, 3.7.1

• Loads for the comparative tests

  One stable load of about Max/3 (eccentricity) and one of 50–100 % of Max (repeatability)

  They don't need to be calibrated unless you also use them for the error test.

  EURAMET cg-18, §5.1, §5.3

• Environment

  Stable temperature, no drafts or vibration, relative humidity 40–60 %

  NIST suggests about ±2 °C per day for general weighing and ±1 °C per day for high precision. Below 40 % RH static electricity causes erratic readings.

  NIST GMP 10, §1

• Thermometer (and barometer)

  Air temperature near the scale, recorded at least once

  Pressure or altitude helps with air buoyancy on high-resolution balances.

  EURAMET cg-18, §5.4

• Paperwork

  Scale identification (Max, d, e, class), tolerance, procedure and record form

  Decide the tolerance before you start. See how to write a calibration procedure.

Calibrate the scale where it is used. Moving it afterward can change its readings through a different local gravity, a different environment or the transport itself, and cg-18 says a certificate shouldn’t be accepted as evidence of traceability for a scale moved after calibration unless that scale has been shown to be immune to those effects (cg-18, §4.1.2).

Which test weights do you need?

Start from the scale’s accuracy class. NIST Handbook 105-1 gives the weight classes that are likely to be suitable for each class of scale (HB 105-1, Table 1):

The table is a starting point. Check the actual numbers at the loads you will test, because the tightest requirement is usually at the top of each tolerance band (HB 105-1, §4). Two examples with OIML R 111 maximum permissible errors (R 111, Table 1):

• A Class III bench scale, 15 kg × 5 g, tested at 15 kg: the maintenance tolerance is 3 e = 15 g, so the weights must be within 5 g. A 10 kg and a 5 kg weight of class M1 are allowed 500 mg and 250 mg, far inside that.
• A Class II balance tested at 5 g with e = 1 mg: the maintenance tolerance is 1 e = 1 mg, so the weight must be within 0.333 mg. A class F1 5 g weight (0.16 mg) is acceptable (HB 105-1, §4, note).

For high-resolution balances, weights can be used with their certified (conventional) values instead of their nominal ones, as in EURAMET’s worked example, and cg-18 adds their drift since calibration and air buoyancy to the uncertainty (cg-18, §7.1.2). ASTM E617-23 has ten classes, 000, 00 and 0 to 7 (ASTM E617-23, 1.1); OIML R 111 has E1 to M3. Handle small weights with forceps or clean gloves, never bare hands (NIST GMP 10, §2).

Test loads and the three tests

EURAMET cg-18 normally uses three tests: repeatability, errors of indication and eccentricity (cg-18, §5). The client and the lab can agree on details and extra tests, but not on fewer than the minimums below.

Errors of indication (linearity)

At least five test loads spread fairly evenly over the range, for example zero or the minimum load, 25 %, 50 %, 75 % and 100 % of capacity. Each actual load may differ from its target by up to 10 % of Max, as long as consecutive loads are at least 20 % of Max apart. For a much smaller agreed range you can use as few as three points, including its minimum and maximum, with consecutive loads no more than 15 % of the range’s upper limit apart (cg-18, §5.2). The test loads must be standard weights (or substitution loads made with them, for large scales). This is the test that shows linearity: whether the error changes along the range.

The loads are normally applied once, in one of four ways: increasing with unloading in between (closest to most real use), increasing continuously, increasing and then decreasing (the sequence used in OIML verification tests), or decreasing from Max (cg-18, §5.2). Loading up and then down is how you see hysteresis: a scale that reads differently at the same load on the way down. Handbook 44’s test notes also include a decreasing-load test for automatic-indicating commercial scales, at the top load of each tolerance band (2.20, N.1.2).

Repeatability

The same load, preferably one single body, placed on the pan at least five times (at least three for loads of 100 kg or more), with the scale zeroed beforehand and the zero checked after each removal. A load between half of Max and Max is common (cg-18, §5.1). The result is the standard deviation of the readings (cg-18, §6.1). You can combine it with the error test by repeating one of the loads. USP <41> (in the text in force until January 2026) asked for at least ten weighings, so pharmaceutical procedures often use ten.

Eccentricity (off-center loading)

A load of about one third of Max, placed in the center and then at four off-center positions (cg-18, §5.3):

Zero the scale, put the load in the center, then move it to the other four positions in any order and record each reading. If the scale drifts, place and remove the load at each position instead, or return to the center between positions (cg-18, §5.3, methods 3 and 4). Platforms with more than four supports, hoppers and vehicle scales need other patterns (OIML R 76-1, 3.6.2; Handbook 44, N.1.3).

Step-by-step scale calibration procedure

The sequence follows EURAMET cg-18 plus the as-found and as-left logic of any calibration. Adapt it to your scales and write your choices into your calibration procedure.

1. 1

   Identify and inspect

   Record ID, model, Max, d (and e and accuracy class if marked), location and the tolerance you will apply. Check the pan, the display and the level bubble for damage or dirt.

   Damaged or contaminated: record as found, do not calibrate (cg-18, §4.1.3)

2. 2

   Prepare

   Check that the site is suitable (drafts, vibration, stable bench). Level the scale. Switch it on for its warm-up time. Let the weights reach the temperature of the room; class E2 and F1 weights are sensitive to this.

3. 3

   Exercise

   Load the scale at least once to about the largest test load; repeated loading is advised (cg-18, §4.1.3). Do not record these readings.

4. 4

   As-found tests

   Run the repeatability, eccentricity and error-of-indication tests. Read only when the indication is stable and record the original readings, not the errors (cg-18, §4.4.1). Record the air temperature.

5. 5

   Evaluate

   Calculate the standard deviation, the largest eccentric deviation and the error at each load, and compare them with your tolerance.

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

6. 6

   Adjust if needed

   Use the manufacturer's method (span or linearity adjustment, internal weight) and only past the threshold in your procedure.

7. 7

   As-left tests

   After any adjustment, repeat the tests and record them as a separate set. If nothing was adjusted, the as-found results are also the as-left results.

8. 8

   Uncertainty, record and label

   Evaluate the uncertainty of each error (or apply your validated values), issue the certificate or record, and label the scale.

Read the indication only when it is stable. If the scale never settles, record an average and note how much the readings varied (cg-18, §4.4.1). On scales where the display step is large compared with the errors you are looking for, labs either switch to a higher-resolution service mode or add small weights to find the exact point where the display changes (cg-18, §4.4.2).

Acceptance criteria and measurement uncertainty

Worked example 1: a 220 g analytical balance (EURAMET cg-18, example H1)

EURAMET’s guide includes a complete worked example: a balance with Max = 220 g and d = 0.1 mg, calibrated with class E2 weights and not adjusted immediately before the calibration (cg-18, App. H1). The guide may only be reproduced in full, so we don’t copy its tables here; the full guide is free from EURAMET, and these are the figures that matter:

• Repeatability: a 100 g load applied five times gave a standard deviation of s = 0.00011 g, a little over one display step (App. H1.2).
• Eccentricity: with the same 100 g load, the largest deviation from the center reading was 0.2 mg (App. H1.2).
• Errors of indication: five loads from 0 to 220 g, each applied once and increasing. The error grew steadily with the load, to +1.3 mg at 220 g: a typical small sensitivity error (App. H1.2).
• Uncertainty: the expanded uncertainty of the error at 220 g is 3.94 mg (k = 2, 95.45 %). At zero the guide uses k = 2.87 instead of 2, because with only five repeatability readings that term has few degrees of freedom (App. H1.3).

The uncertainty is larger than the error at every load, because the guide used the most conservative estimate of air buoyancy: the air density was not known either when the balance was last adjusted or during the calibration. With an estimate based on a 5 K temperature variation during use instead, the uncertainty at 220 g drops from 3.94 mg to 0.98 mg (cg-18, App. H1.3). In this example, the buoyancy estimate matters more than the balance itself.

What goes into the uncertainty

EURAMET cg-18 splits the uncertainty of each error into the uncertainty of the indication and the uncertainty of the reference mass, u²(E) = u²(I) + u²(m_ref) (cg-18, §7.1):

At 220 g in EURAMET’s example, the indication terms together come to about 0.18 mg, while the worst-case buoyancy term alone is about 1.96 mg (cg-18, App. H1.3). Our uncertainty calculator combines a budget like this one.

The guide also separates this uncertainty at calibration from the uncertainty of a weighing in everyday use, which adds the effects of single readings, temperature changes, drift of the adjustment between calibrations and the way the balance is used. Estimates of in-use uncertainty that the lab did not measure may be given with the certificate, but clearly separated from the calibration results (cg-18, §7.4).

Pass or fail

A calibration certificate reports errors; deciding whether the scale is acceptable needs a tolerance. EURAMET cg-18 allows conformity to be declared when the error plus the expanded uncertainty of a weighing result in normal use is within the tolerance, |E| + U(W) ≤ Tol (cg-18, §7.5.3). That in-use uncertainty is not the same number as the calibration uncertainty above. In EURAMET’s example the balance adjusts itself with its built-in weight when the temperature changes, and the guide’s “global” in-use uncertainty, which already includes the error, works out to about 2.8 mg at 220 g (cg-18, App. H1.4): so the balance conforms at that load to a tolerance of ±3 mg, but not to one of ±2.5 mg. Other decision rules exist; our guard banding guide explains them, and whichever you use should be agreed and written in your procedure.

Worked example 2: a commercial bench scale, as found and as left

The second example is fictitious: bench scale SC-07, Handbook 44 Class III, Max 15 kg, e = d = 5 g (n = 3000), tested at five loads before and after adjustment, against the Handbook 44 maintenance tolerance (Table 6). The tolerance steps up with the load: 5 g up to 500 e (2.5 kg), 10 g up to 2,000 e (10 kg) and 15 g above.

• As found: the scale read high, increasingly so with the load: +20 g at 11 kg and +25 g at 15 kg, both outside the ±15 g tolerance. In service, that is a scale that overcharges customers on heavy items.
• As left: after a span adjustment the largest error is +5 g, inside the maintenance tolerance and also inside the acceptance tolerance (half of it, Handbook 44 T.N.3.2) that an official would apply when testing the scale within 30 days after corrective service (G-T.1).

The eccentricity test, with a 5 kg load (Max/3), was fine before and after: no off-center reading differed from the center by more than one division, well inside the 10 g tolerance at that load.

The display step here equals e, so each reading hides up to half a division of rounding. Official tests and careful calibrations remove it with small extra weights (OIML R 76-1, 3.5.3.2; cg-18, §4.4.2).

Minimum weight and USP <41>

A scale’s relative uncertainty grows as the load gets smaller, because parts of its uncertainty, such as repeatability and resolution, don’t shrink with the load. The minimum weight is the smallest net sample for which the relative uncertainty still meets your requirement (cg-18, App. G). It refers to the sample, not the sample plus its container, and users often apply a safety factor because real weighing conditions are worse than at calibration (cg-18, App. G). For EURAMET’s example balance, with a 1 % requirement and a safety factor of 3, the guide gets a minimum weight of 0.0729 g (cg-18, App. H1.4).

USP General Chapter <41> Balances sets this for balances used in pharmaceutical work for materials that must be accurately weighed. A revised chapter became official on February 1, 2026 (see the box below). The text in force until January 2026, as amended in 2014, has two tests (USP <41>, USP 38 text; Brady Systems):

• Repeatability: weigh one test weight at least ten times. Twice the standard deviation divided by the smallest net weight you intend to weigh must not exceed 0.10 %. If the standard deviation is smaller than 0.41 d (d = scale interval), use 0.41 d instead. The test weight doesn’t need to be calibrated.
• Accuracy: the reading of a test weight between 5 % and 100 % of capacity must be within 0.10 % of the weight’s value. The weight’s maximum permissible error, or its calibration uncertainty, must be no more than one third of that limit; the chapter refers to ASTM E617 and OIML R 111.

From the repeatability criterion, the minimum weight is 2000 × s, and a review of the revised chapter gives the same formula, net of the tare container (Burgess and McDowall, LCGC 2026). With the standard deviation of EURAMET’s example balance, 0.00011 g, that is about 0.22 g; under the pre-2026 text, if the balance repeated better than 0.41 d = 0.041 mg, the floor would be 2000 × 0.041 mg = 82 mg. (EURAMET’s example used five repeatability readings; USP asks for ten.)

Handbook 44 has a similar, simpler idea for commercial scales: a recommended minimum load, such as 20 d for Class III scales and 100 d for Class I (Handbook 44, 2.20 UR.3.1, Table 8).

What to record on a scale calibration certificate

Whether you issue a certificate or fill in an internal record, keep (cg-18, §8):

• Identification: scale ID, manufacturer, model, type, Max, d (and e, class), location; the lab and the certificate.
• Procedure: date, place, environmental conditions, the method or procedure number, any limited range agreed, the tolerance applied.
• Adjustment: whether the scale was adjusted, internally or externally, and with which weight.
• Equipment: the test weights (class, certificate) and how traceability is established.
• Results: the indications or errors at each load, the standard deviation (and whether it refers to a single reading), the eccentricity results, as found and as left.
• Uncertainty: the expanded uncertainty of each error, the coverage factor and the coverage probability.
• Decision: pass or fail against the stated tolerance and the decision rule; the minimum weight if you determine it.
• Who and when: technician, date, next due date.

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

How often to calibrate a scale

EURAMET cg-18, OIML R 76 and Handbook 44 don’t set a calibration interval for scales used outside legal metrology. Base it on the risk of a wrong weighing, how heavily and where the scale is used, and what past calibrations found; the 2026 revision of USP <41> asks for the same risk-based approach for calibration and performance checks (ECA Academy, 2024). Our calibration intervals guide explains the methods.

Between calibrations, short routine checks catch problems early. USP <1251> describes such performance checks for analytical balances, with a frequency based on risk (Brady Systems; ECA Academy, 2024); a quick check with one certified weight and an occasional repeatability or corner check are the usual building blocks. Recalibrate, or at least recheck, when a scale is:

• moved to another place (cg-18, §4.1.2);
• repaired, overloaded or dropped;
• reading differently from another calibrated scale, or failing a routine check.

Weights need their own calibration. For field standard weights, NIST notes intervals are often one year at first, extended based on stability data up to a limit set by regulations but generally not beyond five years (HB 105-1, §6.4).

When a scale fails calibration

An as-found error outside tolerance means more than “adjust the scale”. Everything weighed on it since its last good calibration may be wrong, by about the error found at the loads used. In the bench-scale example, items above about 10 kg were weighed up to 25 g heavy. Our guide to out-of-tolerance results walks through the impact assessment.

Common mistakes when calibrating a scale

Related guides

• How to write a calibration procedure: the general structure this procedure follows.
• Scale and balance calibration hub: everything on scales in one place.
• TUR calculator: check that your weights are good enough for the scale.
• Uncertainty calculator: build the uncertainty budget.
• Accuracy vs precision: the difference between error and repeatability.
• Out of tolerance: what to do when a calibration fails.
• Calibration intervals: how often to recalibrate.
• Calibration label generator: print the calibration sticker.