Measurement Uncertainty for ISO/IEC 17025 Laboratories: Calculation, Budgets and Reporting

Every measurement result has a shadow: a range of values within which the true result plausibly lies, given everything we know about the measurement process. That shadow is measurement uncertainty — and for accredited testing and calibration laboratories, understanding, calculating, and reporting it correctly is not optional. It is a core technical requirement of ISO/IEC 17025, a matter of scientific honesty, and increasingly a commercial expectation from sophisticated customers who need to make decisions based on your results.

This guide covers the practical foundations: what uncertainty is and is not, what ISO/IEC 17025 actually requires, how to build an uncertainty budget using the GUM framework, the difference between Type A and Type B evaluation, and how the uncertainty reporting requirement plays out across different measurement disciplines. It also covers how laboratory management software can make uncertainty documentation manageable rather than a manual burden.

1. What Measurement Uncertainty Is

Measurement uncertainty is defined in the International Vocabulary of Metrology (VIM, JCGM 200:2012) as a "non-negative parameter characterising the dispersion of the quantity values being attributed to a measurand, based on the information used." In plain language: it is the interval around a measurement result within which the true value of the quantity being measured is expected to lie, with a stated confidence level.

Understanding what uncertainty is not is equally important, because several related concepts are often confused with it:

• Error is the difference between a single measurement result and the true value — a single number, deterministic in principle, often unknowable in practice. Uncertainty is a range, probabilistic in character, and always estimable even when error is not.
• Accuracy refers to how close a result is to the true value — a qualitative concept. Uncertainty quantifies it.
• Precision refers to the repeatability or reproducibility of a measurement process — how close repeated results are to each other. A process can be highly precise but still have large uncertainty if the systematic effects (bias) are poorly characterised.
• Tolerance is a specification — an acceptable range set by a designer, standard, or customer. Uncertainty is a property of the measurement, not of the thing being measured. A measurement result close to a tolerance limit can only be meaningfully interpreted if the measurement uncertainty is known: a result that appears to fail by 0.1 units means something very different if the measurement uncertainty is ±0.05 units versus ±0.5 units.

Why does measurement uncertainty matter for testing and calibration laboratories? Because a result without uncertainty is incomplete. It tells the customer where the result was, but not how confident they should be in that result. For calibration laboratories issuing certificates that underpin the traceability chain for thousands of downstream measurements, the uncertainty of the calibration directly determines the uncertainty of every measurement made with the calibrated instrument. For testing laboratories, uncertainty determines whether a result genuinely demonstrates conformance or non-conformance with a specification — and getting this wrong has real consequences: passed products that should fail, failed products that are actually within specification, and legal exposure when dispute resolution requires demonstrating the reliability of results.

2. ISO/IEC 17025 Requirements for Uncertainty

ISO/IEC 17025:2017 addresses measurement uncertainty in Clause 7.6, titled "Evaluation of measurement uncertainty." The clause structure reflects the different positions of testing and calibration laboratories.

Calibration Laboratories — No Exceptions

For calibration laboratories, Clause 7.6.2 is unambiguous: the laboratory shall evaluate the measurement uncertainty for all calibrations. There is no exception. Every calibration certificate issued by an accredited laboratory must include a statement of the expanded measurement uncertainty, with the coverage factor stated. Calibration certificates that report only a pass/fail against a tolerance, without numerical uncertainty data, do not meet this requirement — regardless of the discipline or the simplicity of the calibration. This is why Clause 7.6 is one of the first areas NABL assessors and NATA assessors examine when reviewing calibration certificate samples.

Testing Laboratories — Conditional Requirement

For testing laboratories, Clause 7.6.3 requires that laboratories identify the contributions to measurement uncertainty. Clause 7.8.6 then specifies when uncertainty must appear on test reports: when it is relevant to the validity or application of the test result; when the customer's instructions require it; or when it affects compliance with a specification limit (i.e., when the result is close enough to a limit that the uncertainty interval straddles the limit).

Clause 7.6.3 also provides a limited exception: when the nature of the test method prevents rigorous uncertainty evaluation, the laboratory shall at minimum identify the contributing components and make a reasonable estimate. This exception is not a blanket exemption — it requires specific justification per method and cannot be used as a default position for an entire laboratory. Accreditation body guidance documents (NATA's Technical Policy TP-005 and NABL's technical discipline-specific documents) provide further specificity on when the exception is considered appropriate and what "reasonable estimate" means in practice.

When Uncertainty Affects Compliance Decisions

The 2017 edition of ISO/IEC 17025 introduced a new requirement at Clause 7.8.6.1: when a statement of conformance is made — for example, "the sample complies with specification X" — the laboratory shall document the decision rule it applied, including how measurement uncertainty was taken into account. This is a significant practical requirement. A binary pass/fail statement based on a test result does not tell the customer whether the laboratory considered uncertainty when making the conformance call. The decision rule must be agreed with the customer beforehand and stated on the report. For results well inside or outside a specification limit, where the uncertainty interval does not approach the limit, uncertainty may not change the conformance call — but the decision rule still applies.

3. The GUM Methodology

The internationally recognised framework for evaluating and expressing measurement uncertainty is the Guide to the Expression of Uncertainty in Measurement (GUM), published as JCGM 100:2008 by the Joint Committee for Guides in Metrology (JCGM). The GUM is the foundational document for uncertainty evaluation in all physical measurement disciplines and is referenced directly in ISO/IEC 17025. Supplementary documents extend the approach: JCGM 101:2008 covers the propagation of distributions using a Monte Carlo method, and JCGM 102:2011 covers extension to any number of output quantities.

For chemical measurement — where classical GUM propagation through a model equation can be difficult to apply — the EURACHEM/CITAC guide "Quantifying Uncertainty in Analytical Measurement" (EURACHEM/CITAC CG4, 4th edition) provides an alternative top-down approach based on method validation data, which is widely used in chemical and food testing laboratories.

The GUM Process

The GUM approach follows a structured sequence:

1. Define the measurand. Specify precisely what is being measured, including the conditions under which the measurement is made. Ambiguity in the measurand definition leads to ambiguity in the uncertainty evaluation.
2. Identify all uncertainty sources. List every factor that can cause the measurement result to vary — from instrument resolution and calibration to environmental conditions, sampling effects, and operator skill. This list forms the basis of the uncertainty budget.
3. Quantify each source. For each identified source, determine a numerical estimate of its contribution to uncertainty — either by statistical analysis (Type A) or by other means (Type B).
4. Convert all contributions to standard uncertainties. Each source, regardless of how it was evaluated, must be expressed as a standard uncertainty u(x_i) — the equivalent of one standard deviation for that source. This step requires knowing or assuming the probability distribution for each source (normal, rectangular, triangular, etc.).
5. Write the measurement model. Express the output measurand Y as a function of the input quantities X_i: Y = f(X₁, X₂, ..., X_n). This is the model equation.
6. Calculate sensitivity coefficients. The sensitivity coefficient c_i for each input quantity is the partial derivative of the model function with respect to that input: c_i = ∂f/∂X_i. The sensitivity coefficient converts an uncertainty in an input quantity to its contribution to uncertainty in the output — accounting for how strongly the output responds to a change in that input.
7. Combine the contributions. Using the law of propagation of uncertainty, the combined standard uncertainty u_c(y) is calculated as the square root of the sum of the squares of each contribution: u_c(y) = √Σ[c_i · u(x_i)]². This assumes the input quantities are uncorrelated. Where significant correlations exist, the GUM provides the extended formula with covariance terms.
8. Calculate expanded uncertainty. Multiply the combined standard uncertainty by the coverage factor k to obtain the expanded uncertainty U = k · u_c(y). State the confidence level associated with the coverage factor used.

The model equation step is where physical understanding of the measurement process becomes essential. A model equation that omits a significant systematic effect will produce an uncertainty budget that is incomplete — and therefore underestimates the true uncertainty. Common omissions include thermal expansion corrections in dimensional measurement, reagent purity in chemical measurement, and load cell calibration uncertainty in mechanical testing.

EURACHEM/CITAC for Chemical Measurement

The EURACHEM/CITAC CG4 guide offers two complementary approaches for analytical chemistry. The first follows the GUM model equation route, building a budget from individual sources. The second uses a "top-down" approach: precision (repeatability and reproducibility) data from method validation or proficiency testing, combined with bias data from certified reference materials, are used directly as the dominant uncertainty components. This approach is practical for routine chemical testing where the measurement model is complex, and it is increasingly accepted by accreditation bodies including NABL and NATA for accredited chemical and food testing.

4. Type A and Type B Uncertainty Evaluation

The GUM distinguishes two methods of evaluating uncertainty components, called Type A and Type B. The distinction is based on how the numerical value is obtained — not on whether the source is "random" or "systematic," which is a common misconception.

Type A Evaluation

Type A evaluation is based on statistical analysis of a series of observations under defined conditions. The standard uncertainty for a Type A component is the experimental standard deviation of the mean (also called the standard error of the mean):

u(x̄) = s / √n

where s is the experimental standard deviation of the individual observations and n is the number of observations. The standard deviation s is calculated as:

s = √[ Σ(x_i − x̄)² / (n−1) ]

The quantity (n−1) is the degrees of freedom ν for this component. Degrees of freedom affect how reliably the standard deviation estimate represents the true population variability: a standard deviation calculated from three observations (ν = 2) is far less reliable than one from twenty observations (ν = 19). When degrees of freedom are limited, the GUM recommends using the effective degrees of freedom (Welch-Satterthwaite formula) across the full budget to determine the appropriate coverage factor k from the t-distribution rather than assuming k=2.

Type A components commonly arise from: in-house repeatability studies run at the time of method validation; ongoing repeatability data from control charts; and reproducibility data from proficiency testing or interlaboratory comparison exercises.

Type B Evaluation

Type B evaluation uses any means other than statistical analysis. The standard uncertainty is derived from prior information — and the method of derivation depends on what information is available.

Common Type B sources and how to treat them:

• Calibration certificate: If the certificate states an expanded uncertainty U at a stated coverage factor k, then the standard uncertainty is u = U/k. If no coverage factor is stated but the certificate states a 95% confidence level with a normal distribution, use k=2.
• Manufacturer specification: If a manufacturer states a specification as "±a" without stating a distribution, the GUM recommends treating it as a rectangular (uniform) distribution, giving a standard uncertainty of u = a/√3. This is a conservative assumption when the actual distribution is unknown.
• Resolution of a digital display: For a digital instrument displaying the last digit with value d, the resolution contribution is treated as a rectangular distribution over the range ±d/2, giving u = (d/2)/√3 = d/(2√3).
• Reference data from handbooks or standard tables: Treat using the stated uncertainty or, if no uncertainty is stated, make a judgement about the plausible range and apply the appropriate distribution.
• Drift between calibrations: If calibration history shows the instrument drifts by a maximum amount a over the calibration interval, treat as a rectangular distribution: u = a/√3.

The conceptual point that the GUM makes clearly is that a well-founded Type B evaluation is not inferior to a Type A evaluation. An engineer who correctly applies a Type B evaluation to a well-characterised calibration certificate uncertainty is performing a valid and reliable uncertainty analysis. The type label describes method, not quality.

5. Building an Uncertainty Budget

An uncertainty budget is a structured table that lists every identified uncertainty source, its numerical estimate, the probability distribution assumed, the standard uncertainty derived, and the sensitivity coefficient — producing each source's contribution to the combined standard uncertainty. The budget is the working document of the GUM process and the primary evidence an assessor reviews when evaluating the laboratory's uncertainty estimation.

Step 1: Identify All Sources

A common systematic approach is to trace the measurement process from sample receipt to result reporting and ask, at each stage, what can cause the result to differ from the true value. Major source categories for most measurement systems include:

• Resolution — the smallest increment readable from the instrument display or output
• Repeatability — short-term variation in repeated measurements under the same conditions (Type A)
• Reproducibility — variation across operators, days, or environmental conditions (Type A or Type B)
• Calibration of the reference standard or instrument used — the uncertainty stated on the calibration certificate (Type B)
• Reference standard or certified reference material (CRM) value — the uncertainty of the assigned value of the reference
• Environmental conditions — temperature, humidity, vibration effects on the measurement (Type A or Type B)
• Operator effects — variation between operators in reading, positioning, or applying the measurement procedure
• Sample preparation and homogeneity — for chemical and microbiological testing, variation introduced during sample processing
• Sampling — where the laboratory is responsible for obtaining the sample, the uncertainty introduced by the sampling process itself

A source not included in the budget is a source whose contribution is implicitly assumed to be zero. Before finalising a budget, the laboratory should review whether any significant source has been omitted, particularly those specific to the discipline (thermal expansion in dimensional work, matrix effects in chemistry).

Step 2: Quantify and Convert to Standard Uncertainties

For each source, determine the numerical value and probability distribution, then calculate the standard uncertainty u(x_i):

• Normal distribution: u = stated standard deviation (or U/k if expanded uncertainty given)
• Rectangular distribution: u = half-width / √3
• Triangular distribution: u = half-width / √6

Step 3: Apply Sensitivity Coefficients

Multiply each standard uncertainty by its sensitivity coefficient c_i. For additive models (y = x₁ + x₂ + ... ), all sensitivity coefficients are 1 and the contributions equal the standard uncertainties directly. For multiplicative models or those involving calibration factors, the sensitivity coefficients differ and must be calculated from the model equation.

Step 4: Calculate Combined Standard Uncertainty

The combined standard uncertainty u_c(y) is the square root of the sum of squares of all contributions — the "quadrature sum":

u_c(y) = √[ (c₁·u₁)² + (c₂·u₂)² + ... + (c_n·u_n)² ]

Contributions are added in quadrature (root sum of squares) rather than linearly because independent random effects partially cancel. Linear addition would be appropriate only if all contributions were perfectly correlated — the worst case — and would grossly overestimate uncertainty in practice.

Step 5: Apply Coverage Factor and Report Expanded Uncertainty

Multiply the combined standard uncertainty by the coverage factor k to obtain the expanded uncertainty U:

U = k · u_c(y)

For most accredited laboratory reports and calibration certificates, k=2 is used, corresponding to approximately 95% confidence for a normal distribution. The reported expanded uncertainty should include: the numerical value of U; the units; the coverage factor used; and the confidence level. A correctly formatted statement reads: "The expanded measurement uncertainty is U = 0.04 mm (k=2, approximately 95% confidence level)."

Rounding: the expanded uncertainty is conventionally reported to two significant figures. The measurement result is then rounded to the same decimal place as the uncertainty.

6. Uncertainty in Different Measurement Types

The principles of uncertainty evaluation are universal, but the dominant sources and practical challenges differ significantly across disciplines. Laboratories working across multiple disciplines — a common position for TIC organisations covering materials testing, chemical analysis, dimensional inspection, and NDT — must maintain separate uncertainty budgets for each measurement type.

Dimensional Measurement and Calibration

In dimensional measurement and instrument calibration, the dominant uncertainty sources are typically: the calibration uncertainty of the reference standard (gauge block, ring gauge, reference thermometer); the resolution of the measuring instrument; repeatability over a series of measurements at the same point; and thermal expansion — particularly the difference in temperature from the reference temperature of 20°C and the difference in thermal expansion coefficients between the artefact and the instrument material. For high-precision dimensional work, the thermal expansion contribution can dominate the budget when temperature is not controlled to ±0.5°C or better. The model equation for a length measurement explicitly includes a thermal correction term, and the uncertainty of the temperature measurement and the expansion coefficients must both be accounted for.

Chemical Measurement

The EURACHEM/CITAC CG4 guide is the principal reference for chemical measurement uncertainty. The dominant sources are typically: the uncertainty of the certified reference material (CRM) used for calibration or recovery checks; method precision (repeatability and reproducibility); recovery (bias); and, for trace-level measurements, blank uncertainty and interference effects. The top-down approach — using reproducibility data from method validation or proficiency testing as the primary precision component, combined with CRM-derived bias correction uncertainty — is practical for routine chemical testing where deriving a full GUM model equation from first principles is burdensome. Laboratories should review whether method precision data is representative of actual operating conditions: a reproducibility study conducted over two weeks under controlled conditions may underestimate the reproducibility seen across a full year of routine operation.

Mechanical Testing

For mechanical testing (tensile testing, hardness testing, impact testing), calibration uncertainty records for load cells, extensometers, and test machines are critical. Key uncertainty sources include: the calibration uncertainty of the force reference standard used to calibrate the test machine; the calibration uncertainty of the extensometer; repeatability of the test result across replicate specimens; specimen preparation variation (dimensional measurement of the specimen cross-section enters the stress calculation, so its measurement uncertainty propagates into the result); grip effects (alignment and slip under load); and, for temperature-controlled tests, the calibration uncertainty of the temperature measurement system. The sensitivity coefficient for specimen cross-section uncertainty in a tensile stress calculation is typically 1/A where A is the cross-sectional area — meaning that a 1% uncertainty in area measurement contributes a 1% uncertainty to the calculated stress.

Non-Destructive Testing (NDT)

Measurement uncertainty in NDT is an area where the GUM methodology is applicable but requires careful thought about what the "measurand" actually is. For ultrasonic thickness measurement, the measurand is a physical dimension and a conventional uncertainty budget applies: calibration block accuracy, instrument resolution, sound velocity variation, coupling effects, and operator technique. For flaw detection and sizing, the measurand is more complex — the detection probability and sizing accuracy depend on the technique, the inspector, and the flaw geometry in ways that are not easily captured in a single uncertainty statement. NDT certification and technique qualification data (probability of detection curves, sizing error distributions) inform the uncertainty estimate for NDT measurements where the output is a flaw characterisation rather than a dimensional value.

Temperature Measurement

Temperature measurement uncertainty budgets must account for: calibration uncertainty of the reference temperature standard and of the thermometer being used; sensor self-heating; immersion depth or insertion effects for contact sensors; radiation effects for radiation thermometers; temperature homogeneity of the measurement environment (particularly relevant in oven and bath calibrations, where the temperature at the sensor location may differ from the temperature at the reference sensor); and the resolution of the temperature display. For temperature-controlled environments used in testing (e.g., controlled temperature testing chambers, autoclaves), the spatial and temporal uniformity of the temperature field may be the dominant contribution to measurement uncertainty for tests performed within that environment.

7. Managing Uncertainty Records with Software

The technical work of building an uncertainty budget is intellectually demanding — it requires genuine understanding of the measurement process. But once the budget is built, the management burden is equally significant: the budget must be stored, linked to the test method it applies to, versioned when the method changes, reviewed when equipment is recalibrated, and used to generate the uncertainty statement that appears on every test report and calibration certificate issued using that method. This is where laboratory management software moves the task from a documentation burden into a controlled, audit-ready system.

OMS Software provides accredited testing and calibration laboratories with a platform designed around exactly these requirements.

Uncertainty budgets can be stored directly within the method record in OMS — so the uncertainty documentation lives alongside the procedure, the equipment assignments, and the acceptance criteria for that method. When a method is revised, the uncertainty budget is part of the controlled document set that goes through the approval workflow, ensuring that outdated budgets cannot remain in circulation after a method change. When an instrument linked to the method is recalibrated and its new calibration certificate changes the Type B contribution from calibration, the system flags the method for uncertainty review, rather than leaving it to chance or individual memory.

On test reports and calibration certificates generated through OMS, the expanded uncertainty statement is drawn directly from the stored uncertainty record for the method and measurement range used. This eliminates the most common source of reporting errors: a technician manually transcribing an uncertainty value onto a certificate and either using the wrong value or omitting it entirely. For calibration management, OMS stores the calibration certificates received from external providers with their stated uncertainties — which can then be referenced directly when building or reviewing uncertainty budgets for the instruments those certificates cover.

For NABL assessment preparation and NATA assessment preparation, the ability to pull all uncertainty budgets, with their version history, linked calibration records, and the test reports that reference them, in a single retrieval rather than across multiple spreadsheets and shared drives, is the practical difference between an assessment that demonstrates systematic uncertainty management and one that demonstrates it exists but is hard to find. Assessors do not just check whether an uncertainty budget is present — they check whether it is linked to what the laboratory actually does, whether it is current, and whether the uncertainty reported on certificates matches what the budget predicts.

Measurement uncertainty training is another area where the OMS platform supports ongoing competency. Training records for technical staff — including uncertainty estimation training, GUM methodology training, and method-specific uncertainty workshops — are maintained in the personnel module, where they contribute to the authorisation record for each staff member working on accredited methods.