Calibration intervals: how often to calibrate

A calibration interval is the scheduled time between two successive calibrations of an instrument (NASA-HDBK-8739.19-3, Appendix A); it can also be counted in hours of use (ILAC G24, §6.4). No standard sets one number for everything. ISO 9001 asks for calibration at specified intervals or before use and leaves the length to you (clause 7.1.5.2), and ILAC G24 states that there is no universal best practice and that no single method suits all equipment (§4.6). You set an initial interval from a risk assessment, then shorten or extend it based on the as-found results of each calibration, so that a target share of instruments (often 80 % to 95 %) is still in tolerance when it comes due. This guide shows how, with sources an auditor will recognize.

What a calibration interval is (and who sets it)

The interval is the time between calibrations. In practice it decides two things: how much you spend on calibration, and how much product sits “in doubt” if the next calibration finds the instrument out of tolerance. Everything measured since the last good calibration may need to be reviewed in that case, so a longer interval means a bigger out-of-tolerance investigation when something goes wrong.

The user sets it. ILAC G24, the international guidance on recalibration intervals published jointly with the legal-metrology organization OIML, puts the responsibility on the lab or organization that uses the equipment: it chooses the method based on its own needs and risk assessment, checks that the method works, and takes the consequences (ILAC G24, §1.2). The decision should be made by technically competent staff, who estimate how long each instrument or group is likely to stay within its limits after calibration (§5.2).

The calibration lab does not decide for you. Labs accredited to ISO/IEC 17025 must not put a recommended interval on the certificate or label unless you agreed to it, although law or regulation can override this rule (clause 7.8.4.3, as summarized in NIST’s crosswalk). And a certificate does not “expire”: it reports results on one date. See our guide to reading a calibration certificate.

Managing an interval is a loop, not a one-time decision:

1. 1

   Set the initial interval

   From a risk assessment: manufacturer data, similar equipment, use, environment, how critical the measurements are.

2. 2

   Calibrate and record the as-found data

   Results before any adjustment, with uncertainty. This is the data the interval is built on.

3. 3

   Compare with the limits

   Was the instrument in tolerance when it arrived?

   Out of tolerance: start an impact assessment

4. 4

   Review the interval with a documented method

   Extend, keep or shorten it, within a minimum and a maximum you have set.

5. 5

   Add intermediate checks where useful

   Short checks between calibrations catch drift early and narrow any future investigation.

6. 6

   Update the record and the schedule

   New due date in the equipment record and on the label; keep the history for the next review.

Is there a standard that says how often to calibrate?

No general one. This is the question people ask most often, usually because management wants a document that says “annual” or “every two years”. Here is what each framework actually requires. For paid standards, we describe requirements in our own words, based on the guidance cited.

So “12 months” is a convention, not a requirement. It can be a reasonable starting point, but you still need to show how you chose it and how you check it.

How to set the initial interval

ILAC G24 bases the initial interval mainly on a risk assessment, and lists fourteen factors to consider, without making the list exhaustive (§5.1): the uncertainty you need, the type of equipment, the risk of exceeding its maximum permissible error, the manufacturer’s recommendation, wear and drift, how much and how hard it is used, the environment, the effect of what is measured (for example thermocouples at high temperature), data from similar equipment, cross-checks against other standards, intermediate checks, transport, staff training, and legal requirements.

When there is no calibration history, NASA’s guidance names the usual bases for a first interval: the manufacturer’s recommendation, intervals used for similar equipment, organizational policy, government directives and contract terms (NASA-HDBK-8739.19-2, §10.2.8 and §11.4). Two cautions:

• A manufacturer’s recommendation is only useful if you know three things: the tolerance limits, how long the instrument is expected to stay within them, and with what probability. Manufacturers may not state all three (NASA-HDBK-8739.19-2, §11.4.2).
• New equipment should be calibrated more often at first, to learn how it behaves before you lengthen the interval (ILAC G24, §6.1).

Some organizations publish initial intervals for specific equipment. They are useful reference points, but each comes from a particular context, so don’t treat them as “the standard interval”:

Use the checklist below to document the reasoning behind each initial interval. That record is what an auditor will want to see.

The interval as a reliability target (EOPR)

Here is the idea that turns interval-setting from guesswork into a method. After calibration, the uncertainty in an instrument’s error always grows with time: the error may drift up, drift down or stay put, but you become less sure of it. That means the probability that the instrument is still in tolerance falls as time passes (NASA-HDBK-8739.19-2, §10.2.8; NASA-HDBK-8739.19-3, §9.1). A good program picks a reliability target and sets the interval where the instrument’s reliability curve reaches it (NASA-HDBK-8739.19-2, §10.2.8, Fig. 10-4).

The probability of being in tolerance at the end of the interval is the end-of-period reliability (EOPR) (NASA-STD-8739.12A, §3.2). In practice you estimate it from your history: the share of calibrations in which instruments of a given type arrived in tolerance.

The chart uses NASA’s own worked example: an instrument that is 98 % likely to be in tolerance right after calibration and 90 % likely after 12 months. If your target were 95 %, the same instrument would need calibration every 8.6 months. If 80 % were acceptable, about every 16 months. The instrument is the same; the target changes the interval.

Which target? NASA’s decision-risk handbook says intervals are commonly set to reach EOPR targets between 80 % and 95 % (NASA-HDBK-8739.19-4, §5.3.3.2). NIST’s GMP 11 ties the target to how critical the standard is: 99 % for standards that contribute more than 25 % of a measurement’s uncertainty, and 95 % for those contributing between 1 % and 25 % (GMP 11, §2.3.2). Two warnings come with high targets:

• With a high target, a few out-of-tolerance results can shorten the interval drastically. That makes false out-of-tolerance calls expensive, because intervals react more to failures than to passes (NASA-HDBK-8739.19-4, §5.3.3.2 and §1.1.2).
• The in-tolerance rate you observe is almost always lower than the true one, because when most instruments are in tolerance, false rejects outnumber false accepts (NASA-HDBK-8739.19-4, §5.3.3.1). Our TUR calculator can correct an observed EOPR for this.

Why the interval changes your false-accept risk

The EOPR is the link between your interval and the risk of passing an instrument that is actually out of tolerance. A shorter interval raises the EOPR, which lowers the probability of false accept (PFA) for the same test uncertainty ratio (NASA-HDBK-8739.19-4, §5.1). Our own calculation with the validated model in our TUR calculator shows this holds above about 70 % EOPR. At a TUR of 4:1 and simple acceptance, the PFA is about 2 % at 79 % EOPR and about 0.9 % at 95 % EOPR.

NASA uses exactly this link. Its 2024 metrology standard accepts an observed EOPR as evidence of meeting the 2 % false-accept limit of ANSI/NCSL Z540.3, provided there are at least 20 calibrations of the same model, with the same procedure and the same interval (NASA-STD-8739.12A, §4.3):

How to adjust the interval: the five ILAC G24 methods

An initial interval is a guess. ILAC G24 lists the reasons to review it: the equipment may be more or less reliable than expected, it may be used differently, and observed drift may call for shorter intervals or allow longer ones without more risk (§6.1). Which review method fits depends on whether you manage instruments one by one or in groups, whether they drift with time or with use, whether they are adjusted at calibration, and how much data you have (§6.1). Whatever you choose, keep the calibration data as a history and document the method (§4.8).

ILAC G24 describes five methods (§6.2–6.6):

1. Staircase (calendar time). At each routine calibration, if the deviation is within a set fraction of the maximum permissible error, the interval is extended (or kept); if not, it is shortened. It reacts fast and needs little administration, but makes workload harder to plan. G24 warns against using it to reach extremely long intervals, because a failure then puts a long stretch of results in doubt (§6.2). G24 gives no percentages or factors: you define them.
2. Control chart (calendar time). Plot significant calibration points over time, estimate scatter and drift, and calculate the interval from them. It needs good knowledge of the instrument and suits single-value standards such as gauge blocks or standard resistors, but not equipment without instrumental drift (§6.3).
3. In-use time. A variant of methods 1 and 2 in which the interval is counted in hours of use. It fits thermocouples at extreme temperatures, standard lamps, deadweight testers and gauges that wear. It does not fit passive standards, or equipment that drifts while idle (§6.4).
4. In-service checking (“black box”). Critical parameters are checked often (daily, for example) with portable equipment or a purpose-built check device, and the instrument goes for full calibration only when a check fails. Availability is excellent, but the check may miss a parameter, and the check device needs its own interval (§6.5).
5. Other statistical approaches. Analyses of individual instruments or of groups, especially when there are large numbers of identical ones, increasingly with the help of software. G24 refers to NCSL International’s RP-1 for detailed examples (§6.6).

G24 compares the five methods in its Table 1 (§6.7):

G24 notes that the statistical methods score better when good software is available. The same methods can also be used to review the type and frequency of intermediate checks (§6, note).

Worked example: adjusting from the last three calibrations

A staircase rule needs numbers, and G24 leaves them to you. One published example comes from Savannah River National Laboratory, a US Department of Energy lab, which presented a simplified algorithm at the NCSL International conference in 2006 (Bare, 2006). It is a conference paper, not a standard, and the author presents the coefficients as an example to tune to your own reliability target.

The new interval is the current interval multiplied by a weighted score of the last three as-found results:

New interval = current interval × (0.8 × X + 0.2 × Y + 0.1 × Z)

where X is the most recent calibration, Y the one before and Z the one before that. Each as-found result scores 1 if in tolerance, 0.8 if out by less than one tolerance band, 0.6 if out by one to two bands, 0.4 if out by two to four, and 0.3 if out by more than four. The weights add up to 1.1, so three clean results extend the interval by 10 %.

We reproduced every result in the paper’s table. The paper also gives three rules worth copying whatever coefficients you use:

• Set a minimum and a maximum interval. Without them, a poor instrument’s interval keeps shrinking toward zero, and a good one’s grows without limit.
• Have an engineer review each proposed change before it takes effect.
• Don’t let the algorithm change intervals that a requirement fixes or caps, such as a maximum cycle.

Extending or shortening an interval without audit trouble

People worry that an extended interval will be written up as a nonconformity. What protects you is the evidence, not the length. NIST’s GMP 11 is explicit about what counts (§2.4.2–2.4.3):

• Good reasons: calibration history, measurement assurance data, interlaboratory comparisons, population data for similar equipment, and recommendations from NIST or the manufacturer, analyzed and documented technically and statistically.
• Not acceptable reasons: not having a system to track due dates, lack of budget, loss of staff, or similar operational pressures.

Cost still matters, but as part of the risk balance. ILAC G24 weighs the cost of more frequent calibration against the larger uncertainty and lower reliability of longer intervals (§4.5). The argument that usually convinces management is the other side of that balance: if the instrument fails at its next calibration, how many months of product do you want to have in doubt?

Intermediate checks are often the better alternative to a shorter interval. ILAC G24 recommends a suitable system of intermediate checks between calibrations (§4.9), and ISO/IEC 17025 asks labs to carry them out by a procedure when they are needed to keep confidence in the equipment (clause 6.4.10, as summarized by A2LA G149 and PJLA). A quick check with a reference artifact catches drift early and, if a calibration later fails, moves the start of the review window forward. NATA’s table, for example, pairs a 2-year CMM calibration with a volumetric check every 6 months.

Recalibrate out of schedule when something happens. An overload, a repair or improper handling (a drop, for example) calls for recalibration whatever the interval (DKD-R 6-1, Appendix F). Check results against predetermined limits before the instrument goes back into service (ILAC G24, §4.10).

After an out-of-tolerance result, two things follow: shorten the interval according to your method, and assess the impact on everything the instrument measured since its last good calibration. The second is covered step by step in our out-of-tolerance guide. Records make all of this possible: see our guide to calibration records.

Common myths about calibration intervals

Next steps: see how your EOPR translates into false-accept risk with the TUR calculator, and how intervals fit into a complete calibration program.