These two words are treated as synonyms in everyday lab talk, and they are not remotely the same thing. One of them you can almost never calculate. The other you can always calculate. Knowing which is which changes how you handle a measurement.
The compressed version:
- Error is the difference between your measured value and the true value. To find it, you need to know the true value.
- Uncertainty is the range the true value probably lies in. You work it out from your own data, without ever knowing the answer.
And the practical consequence, which is the part worth remembering: you correct for error, and you report uncertainty.
The Short Answer
| Error | Uncertainty | |
|---|---|---|
| Definition | Measured value − true value | Range the true value probably lies in |
| Needs the true value? | Yes | No |
| Can you calculate it in routine work? | Almost never | Always |
| Has a direction? | Yes positive or negative | No symmetric band |
| Is it a single number? | Yes, in principle | Yes, but as a ± interval |
| What you do with it | Correct for it, if known | Report it |
| Required by ISO/IEC 17025? | No | Yes |
The second row is the whole story. Error is defined in terms of something you usually can’t access. That single limitation is why modern metrology is built around uncertainty instead.
What Is Measurement Error?
Measurement error is the difference between a measured value and the true value of the quantity.
Error = measured value − true value
Measure a certified 50.00 °C reference and read 50.31 °C, and your error is +0.31 °C. Straightforward because someone certified the reference.
Now measure an unknown sample and read 62.47 °C. What’s your error?
You have no idea. There’s no reference to compare against. The error certainly exists your reading is not exactly right but its value is inaccessible to you.
That’s the fundamental problem with error as a working concept. It’s perfectly well defined and almost never computable.
What Is Measurement Uncertainty?
Uncertainty is the range of values within which the true value probably lies.
Result: 62.47 ± 0.03 °C
You built that ±0.03 °C from things you can observe and document: how much your readings scattered, what your instrument’s resolution is, what its calibration certificate says, how far the temperature strayed from standard conditions.
Notice what you didn’t need: the true value. Uncertainty is calculable precisely because it never refers to the answer. It refers to your process.
That’s why every accredited laboratory in the world reports uncertainty rather than error. It’s the statement you can actually make.
The Core Difference: One Needs the True Value
Everything else follows from this.
Error is retrospective and comparative. It asks: how far off was I? You can only answer with a reference.
Uncertainty is prospective and self-contained. It asks: how far off could I plausibly be? You answer from your own evidence.
There’s a second difference worth noticing. Error has a sign. +0.31 °C means you read high. Uncertainty has no direction it’s a symmetric band either side of your result, because you don’t know which way you’re off.
And a third: error is a property of a particular measurement. Uncertainty is a property of the measurement process, which is why it can be stated in advance, before you’ve measured anything.
Types of Measurement Error
Error is traditionally split by cause. Three categories, and they behave completely differently.
Systematic Error
Shifts every reading the same way by roughly the same amount. Repeating won’t reveal it.
- Zero error on a micrometer or balance
- An instrument reading consistently high
- Glassware calibrated at 20 °C used at 30 °C
- Always reading a scale from the same wrong angle
These are the ones you correct. A known systematic error should be subtracted, not endured.
Random Error
Scatters unpredictably sometimes high, sometimes low.
- Reaction time on a stopwatch
- Small fluctuations in measuring force
- Electrical noise
- Judging an endpoint colour change
These you average down. Random error partly cancels across repeated readings, which is why the mean is more reliable than any single value.
Gross Error
Mistakes. Misreading a scale, transposing digits, using the wrong formula.
These you find and discard. Gross errors aren’t uncertainty and shouldn’t be averaged into anything. If one reading is wildly out of line, check your working rather than including it.
Absolute, Relative and Percentage Error
When you do have a reference value, these are the formulas.
Absolute error:
Absolute error = |measured value − true value|
|50.31 − 50.00| = 0.31 °C
Relative error:
Relative error = absolute error ÷ true value
0.31 ÷ 50.00 = 0.0062
Percentage error:
Percentage error = (absolute error ÷ true value) × 100
0.0062 × 100 = 0.62%
One caution that costs marks constantly. Percentage error and percentage uncertainty are different quantities. Percentage error compares your result to an accepted value. Percentage uncertainty describes the spread in your own measurement. They usually have very different magnitudes, and a question asking for one will not accept the other.
How Uncertainty Is Quantified Instead
Uncertainty is built rather than compared. The process:
- List every source of doubt — instrument, reference standard, environment, operator, workpiece, method.
- Convert each to a standard uncertainty. Repeated readings give s ÷ √n. A certificate value is divided by its k. A tolerance band is divided by √3.
- Combine in quadrature: u꜀ = √(u₁² + u₂² + …)
- Expand: U = k × u꜀, with k = 2 giving about 95% confidence.
- Report as value ± U, stating k.
Nowhere in that sequence does a true value appear. Every input comes from your equipment, your data, or a document.
The Error Approach vs the Uncertainty Approach
This is the part almost no article covers, and it explains why the vocabulary changed.
Until the early 1990s, measurement science used what’s now called the error approach. It assumed:
- A true value exists and is, in principle, discoverable
- Every result equals the true value plus an error
- That error splits into systematic and random components
- The job is to estimate both and correct them
The GUM replaced this with the uncertainty approach:
- The true value is unknowable, and pretending otherwise is unhelpful
- A result is a best estimate accompanied by a stated interval
- No attempt is made to determine “the error”
- Contributions are classified by how they were evaluated (Type A or Type B), not by what caused them
Why the change?
Because the random/systematic split turns out to be unstable it depends on how you draw the boundary around your measurement.
Take an instrument that reads 0.02 mm high. For you, using that one instrument, this is a systematic error. But look at a population of a hundred such instruments, each with its own small offset, and those offsets scatter randomly. The same physical effect is systematic at one scale and random at another.
Type A and Type B don’t have that problem. They describe how you obtained the number statistically from repeated observations, or by other means. That’s unambiguous regardless of where you draw the boundary.
This is why modern certificates state uncertainty with a coverage factor, and don’t mention systematic or random error at all.
Why “True Value” Is a Problem
Worth pausing on, because it’s the root of everything above.
The true value of a quantity is, by definition, the value you’d get from a perfect measurement. Since no perfect measurement exists, the true value is unknowable in principle not just in practice.
Even a certified reference doesn’t escape this. A gauge block certified at 25.0004 mm carries its own uncertainty, because it was measured by an instrument that was itself calibrated against something else. Follow the chain back to a national metrology institute and you’ll find uncertainty at every link.
So when you calculate an “error” against a certified standard, you’re really calculating the difference between your result and someone else’s result with a smaller uncertainty. It’s useful. It’s just not the truth.
The VIM handles this by defining error in terms of a reference quantity value rather than a true value. A pragmatic fix for a concept that doesn’t survive close inspection.
What You Do With Each: Correct vs Report
The practical rule, and the one most people get wrong.
A known error should be corrected.
If your certificate says the instrument reads 0.32 °C high, subtract 0.32 °C from your readings. That’s information you paid for use it.
Uncertainty should be reported.
After correcting, state the residual doubt as ± a value at a stated coverage factor.
Do not absorb a known error into your uncertainty.
This is the common mistake. Widening your uncertainty band to swallow a bias you already know about is doubly wasteful: you throw away a correction you have, and you report a worse result than you achieved.
What does enter your uncertainty is the doubt about the correction the certificate’s own uncertainty, and any drift since it was issued.
Error in Physics and Chemistry vs Uncertainty in Metrology
Here’s a source of genuine confusion that nobody explains: your textbook and a calibration certificate use different words for overlapping ideas.
School and undergraduate syllabuses NCERT, A-Level, IB are built around error vocabulary. Professional metrology, ISO/IEC 17025 and the GUM are built around uncertainty vocabulary. Rough translation:
| School / lab-course term | Metrology equivalent |
|---|---|
| Absolute error | Absolute uncertainty (loosely) |
| Mean absolute error | A Type A evaluation |
| Relative error | Relative standard uncertainty |
| Random error | Contributions typically evaluated as Type A |
| Systematic error | Corrected bias, plus Type B contributions |
| Least count uncertainty | A Type B contribution, rectangular |
They’re not exact equivalents, and one pair is a genuine trap.
Percentage error in a school lab usually means deviation from the accepted value comparing your g against 9.8 m/s², say. That is genuine error, not uncertainty. Percentage uncertainty means something else entirely. If a question asks for one and you give the other, the numbers won’t even be close.
If you’re moving from a physics or chemistry course into engineering or laboratory work, expect the vocabulary to shift under you. The underlying maths barely changes.
Worked Example: Both From the Same Measurement
The clearest way to see the difference is to extract both from one dataset.
The situation: checking a digital thermometer against a certified reference at 50.00 °C. Five readings:
50.31, 50.34, 50.30, 50.33, 50.32 °C
Calculating the error
Mean = 251.60 ÷ 5 = 50.32 °C
Systematic error (bias) = 50.32 − 50.00 = +0.32 °C Error of the first reading = 50.31 − 50.00 = +0.31 °C Random component of the first reading = 50.31 − 50.32 = −0.01 °C
Every one of those numbers required the certified reference. Without it, none is computable.
Calculating the uncertainty
Sample standard deviation, s = 0.016 °C
| Source | Standard uncertainty |
|---|---|
| Repeatability (0.016 ÷ √5) | 0.007 °C |
| Reference thermometer certificate (0.02 at k=2) | 0.010 °C |
| Resolution (0.01 °C) | 0.003 °C |
u꜀ = √(0.007² + 0.010² + 0.003²) = 0.013 °C U = 2 × 0.013 = 0.03 °C
Putting them together
| Quantity | Value | Computable without a reference? |
|---|---|---|
| Systematic error | +0.32 °C | ❌ No |
| Error of one reading | +0.31 °C | ❌ No |
| Random component | −0.01 °C | ❌ No |
| Expanded uncertainty | ±0.03 °C | ✅ Yes |
Corrected result: 50.00 ± 0.03 °C (k = 2)
What this shows
The error was 0.32 °C. The uncertainty is 0.03 °C. The error is more than ten times larger.
That’s the normal relationship, and it’s the reason correction matters so much. A known bias is usually far bigger than the residual doubt, and removing it is the single most valuable thing you can do with a calibration certificate.
And once this thermometer goes off to measure an unknown sample, you can still state ±0.03 °C. You will never again be able to state the error.
Where Error Still Matters
The uncertainty approach didn’t make error obsolete. It’s still the right concept in three situations:
Calibration. The entire purpose is measuring a known artefact to determine the instrument’s error, so it can be corrected. No error, no calibration.
Verification against a specification. Checking whether an instrument’s error stays within its maximum permissible error is a legitimate, common test.
Teaching. The random/systematic distinction is genuinely useful for understanding what is going wrong with a measurement, even though it’s no longer how uncertainty is classified. Knowing that averaging fixes one and not the other is practical knowledge.
Error explains causes. Uncertainty quantifies doubt. They answer different questions.
Common Misconceptions
“Error and uncertainty are the same thing.” Error needs the true value. Uncertainty doesn’t. In routine work you can calculate one and not the other.
“A large uncertainty means a large error.” Not necessarily. A measurement with a wide uncertainty band might happen to sit very close to the true value. Uncertainty describes what you know, not how wrong you are.
“Zero error means no error.” “Zero error” is the name of a specific fault an instrument reading non-zero when it should read zero. It’s the presence of a systematic error, not the absence of error.
“You can eliminate error by repeating the measurement.” Repeating reduces random error. Systematic error survives untouched, and averaging makes it look more convincing.
“Percentage error and percentage uncertainty are interchangeable.” They’re different quantities with different formulas and usually different values.
“If I don’t know the true value, I can’t say anything about my measurement.” The opposite. That’s precisely the situation uncertainty was designed for.
“A known bias should be included in the uncertainty.” Correct it instead. Only the residual doubt about the correction belongs in the budget.
Frequently Asked Questions
What is the difference between measurement uncertainty and error?
Error is the difference between a measured value and the true value, so you need the true value to calculate it. Uncertainty is the range within which the true value probably lies, calculated from your own data without needing the answer.
Can you calculate error without knowing the true value?
No. Error is defined in terms of the true value. In practice a certified reference stands in for it, which is why error is mainly computed during calibration.
What is the formula for measurement error?
Error = measured value true value. Absolute error takes the modulus; percentage error is (absolute error ÷ true value) × 100.
Is uncertainty the same as percentage error?
No. Percentage error compares your result to an accepted value. Percentage uncertainty describes the spread in your own measurement. Different formulas, usually very different values.
What are the three types of measurement error?
Systematic error, which shifts every reading the same way; random error, which scatters unpredictably; and gross error, which is a mistake and should be discarded rather than analysed.
Why did metrology move from error to uncertainty?
Because the true value is unknowable, and because the random/systematic split depends on how you define the measurement. Type A and Type B classify contributions by how they were evaluated, which is unambiguous.
Should I correct for error or include it in my uncertainty?
Correct for it. A known bias should be subtracted, with only the residual doubt about the correction entering the uncertainty budget.
Does uncertainty include systematic error?
Not a known and corrected one. It includes the doubt about that correction, plus any systematic effects you’ve quantified but not removed.
Which does ISO/IEC 17025 require?
Uncertainty. Accredited laboratories must evaluate and report measurement uncertainty. Error appears only in the context of calibration and verification against specifications.
Do physics and chemistry courses use “error” incorrectly?
Not incorrectly differently. School syllabuses are built on the older error framework, which remains a valid teaching approach. Professional practice uses the uncertainty framework. Expect the vocabulary to shift when you move between them.
Conclusion
Two concepts that overlap in conversation and diverge completely in practice.
- Error = measured value − true value. It has a sign, and it requires knowing the answer.
- Uncertainty = the range the true value probably lies in. It’s symmetric, and it requires only your own evidence.
- You correct for error. You report uncertainty. Never absorb a known bias into your uncertainty band.
- Error is usually much larger than uncertainty ten times larger in the worked example above which is exactly why correction is worth doing.
- Modern metrology is built on uncertainty because the true value is unknowable and the random/systematic split shifts depending on where you draw the boundary.
- Error still matters in calibration, in verification against specifications, and as a teaching tool for understanding causes.
