Why a perfectly repeatable measurement can still be completely wrong.
People use "accurate" and "precise" as if they were the same compliment. In measurement and quality engineering they describe two completely independent things, and a gage, an instrument, or a process can score high on one and low on the other. Confusing them is one of the most consequential mistakes in metrology — it's how a mis-calibrated gauge keeps passing bad parts with total confidence.
Accuracyasks: how close is a measurement to the true value? It's about correctness — bias, offset, error from the real answer, however that real answer is established (usually a calibrated reference standard).
Precision asks a completely different question: how close are repeated measurements to each other? It says nothing about whether they're close to the true value — only about how tightly they cluster. Precision is also called repeatability. A process can repeat itself with beautiful consistency while being consistently wrong.
Each panel plots repeated measurements of the same quantity around the true value (center crosshair).
The top-right panel is what makes the point: those six readings are tightly clustered — genuinely excellent repeatability — and every single one is wrong. That's not a contradiction. Precision is purely a statement about spread (variance, standard deviation). Accuracy is purely a statement about the distance of the average reading from the truth (bias). A tight cluster can sit anywhere; nothing about tightness forces it to sit on target.
Precision is essentially the standard deviation of repeated readings — a measure of random error / noise, and it can be excellent even when every reading shares the same systematic error. Accuracy is essentially the distance between the averageof those readings and the true value — a measure of bias, usually introduced by a miscalibrated zero point, a worn sensor, thermal drift, or a stale reference. Because one measures spread and the other measures offset, improving one does nothing to fix the other: tightening repeatability further won't correct a bias, and correcting a bias won't tighten a scattered process. That's exactly why calibration (comparing against a traceable reference standard) and repeatability studies (like Gage R&R) are two separate disciplines that both matter.
No — that's precision, not accuracy, and the pressure gauge above is the textbook counter-example. A gauge with zero-offset drift will reproduce the same wrong answer with total consistency, run after run, because whatever is causing the offset (a sagged spring, a shifted zero, an uncorrected span) is a fixed, repeatable error — it doesn't introduce scatter, it introduces bias. Repeatability tells you the instrument agrees with itself. Only comparison against a traceable calibration reference tells you whether it agrees with reality. This is precisely why calibration programs check instruments against certified reference standards on a schedule, rather than simply confirming that the instrument reads consistently — consistency is necessary for confidence in a measurement, but it is nowhere near sufficient for correctness.
Accuracy and precision describe two independent properties of a measurement process. Accuracy is how close a reading is to the true value (a question of bias); precision is how close repeated readings are to each other (a question of spread). An instrument can be precise without being accurate, accurate without being precise, both, or neither — and confusing the two is one of the most common and most consequential errors in measurement and quality engineering.
Precision is captured by the spread of repeated measurements — typically the standard deviation of a sample of readings taken under the same conditions (repeatability), or across operators/instruments (reproducibility). Accuracy is captured by the difference between the mean of those measurements and the true or reference value — the bias. A process can have near-zero variance (highly precise) while its mean sits far from the truth (highly inaccurate), because variance and bias are mathematically independent quantities; reducing one has no mechanical effect on the other.
The most common real-world cause is calibration drift: a zero-offset shift, span error, sensor aging, or thermal effect that shifts every reading by roughly the same fixed amount. Because the error is systematic rather than random, it shows up as a consistent offset rather than as scatter — the instrument keeps agreeing with itself while disagreeing with reality. This is why repeatability alone (checking that an instrument reads the same value twice) can never substitute for calibration (checking that it reads the correct value against a traceable reference standard).
This distinction underlies measurement system analysis (Gage R&R, which separates repeatability and reproducibility from bias/linearity studies), statistical process control (a process can be in control — stable and predictable — while still centered off its target), and any calibration program governed by a quality standard such as ISO 9001 or ISO/IEC 17025. A production line running a precise-but-biased gauge will consistently pass or fail parts incorrectly with complete internal confidence, which is exactly why calibration intervals and traceable reference standards are mandatory in regulated manufacturing.
Yes. If repeated readings scatter widely but their average lands close to the true value, the process is accurate on average but not precise — you can trust the long-run mean, but not any single reading. This is common with noisy sensors or manual measurement techniques with inconsistent technique.
Precision is generally expressed as a standard deviation, variance, or repeatability/reproducibility figure (as in a Gage R&R study). Accuracy is generally expressed as bias or percent error relative to a certified reference value. Total measurement error combines both.
Recalibrate it against a traceable reference standard — adjusting the zero offset and/or span so the mean of its readings matches the reference. Recalibration corrects bias; it does not need to touch whatever is already producing good repeatability, since that mechanism is unrelated to the offset.
It depends on the use. For pass/fail inspection against a tight specification, accuracy (correct on average) usually matters most, since a biased gauge systematically misjudges parts. For process control and detecting shifts over time, precision matters more, since a stable, repeatable signal makes small real changes easier to detect even if there is a known, correctable offset.
A Gage R&R study specifically isolates precision into repeatability (variation from the same operator/instrument on repeated trials) and reproducibility (variation between different operators or instruments). Accuracy — bias and linearity against a reference standard — is assessed separately, typically in a calibration or linearity study, because the two properties require different corrective actions.
Try our Industrial & Systems Engineering Studio
More calculators, simulators, and guides for this discipline.