Accuracy vs Precision in Measurement

By Wu Peng, Senior Process Instrumentation Engineer · Last reviewed July 27, 2026

Accuracy is how close a measurement is to the true value. Precision is how close repeated measurements are to each other. The two are independent: a pressure transmitter can repeat the same wrong number all day (precise but not accurate), and another can scatter readings evenly around the right answer (accurate on average but not precise). On an instrument datasheet, accuracy is the published error band; precision shows up as the repeatability line.

The difference matters when you buy and when you calibrate. Bias can be trimmed out during calibration; poor repeatability cannot, which makes repeatability the hard floor of what an instrument will ever do. This guide gives the definitions, a worked five-reading example on a 0–10 bar transmitter, what the terms look like on a real spec sheet, and which one to pay for in control, billing, and alarm service.

Contents

• Definitions
• A worked example
• Repeatability and reproducibility
• Accuracy on datasheets
• Calibration removes bias
• Resolution is different
• Choosing the right class
• FAQ

Definitions

Accuracy answers one question: how far the reading sits from the true value. It is usually quoted as a percent error against a reference standard:

Percent error = (measured − true) / true × 100

Precision answers a different question: how tightly do repeated measurements of the same thing group together. The usual metric is the standard deviation of the repeated readings. Small standard deviation means high precision, regardless of whether the group sits on the true value or not.

Because the two are independent, every instrument lands in one of four boxes:

Combination	What the readings look like
Accurate and precise	Tight group centered on the true value. This is what you pay for.
Precise, not accurate	Tight group offset from the true value. Classic zero shift or span error; usually correctable by calibration.
Accurate, not precise	Wide scatter whose average happens to sit on the true value. Any single reading may still be far off.
Neither	Wide scatter, offset center. Time to repair or replace.

Textbooks draw this as darts on a target. Reading sets from a real transmitter tell the same story with numbers you can act on.

A worked example

Take a gauge pressure transmitter on a test bench, with a deadweight tester holding a true 6.00 bar. Two units, five readings each:

Unit	Readings (bar)	Mean	Std dev
Unit A	6.13, 6.14, 6.14, 6.14, 6.15	6.14	0.007 bar
Unit B	5.96, 6.05, 5.99, 6.03, 5.97	6.00	0.039 bar

Unit A is the precise one: its five readings sit within 0.02 bar of each other. But every one of them is about 0.14 bar high, a bias of 2.33 percent of the true value. Unit B looks worse at a glance, yet its mean lands exactly on 6.00 bar; it is accurate on average with a standard deviation five times larger than Unit A.

Between the two, I would rather own Unit A. Its error is a stable, repeatable offset, and a zero trim against a reference removes almost all of it. Unit B has nothing to trim; its scatter is random, and no adjustment can shrink it.

To put the numbers in loop terms: on a 0–10 bar span transmitted over 4–20 mA, an accuracy band of ±0.5 percent of full scale is ±0.05 bar, which is ±0.08 mA at the output. Unit A’s 0.14 bar bias would sit well outside that band until you trim it.

Repeatability and reproducibility

Precision splits into two layers, and standards bodies keep them separate for good reason.

Repeatability is the spread you get when the same instrument, the same operator, and the same conditions produce readings over a short period. That is what the five-reading sets above measure, and it is what the repeatability line on a transmitter datasheet promises.

Reproducibility is the spread when something changes between measurements: a different technician, a different day, a different test rig, approach from rising pressure instead of falling. Reproducibility is always equal to or worse than repeatability, because it adds those extra sources of variation on top. Reliability, in the everyday sense of readings you can trust next month, comes from repeatability plus low drift rather than from a tighter accuracy class.

Engineering practice usually states precision as two or three standard deviations. At three standard deviations, about 99.7 percent of readings fall inside the band; Unit A above would carry a three-sigma repeatability of roughly ±0.02 bar. One more term belongs here: long-term drift is neither repeatability nor reproducibility. It is a slow walk of the zero or span over months, and datasheets list it separately as stability, typically in percent per year.

Accuracy on datasheets

Spec sheets rarely use the word precision. Instead the terms appear like this:

Reference accuracy is the headline number, and it is a bundle: by convention it includes non-linearity, hysteresis, and non-repeatability together, stated at reference conditions of roughly room temperature and stable ambient. Our differential pressure transmitters state a reference accuracy of 0.075 percent of span; on a 0–100 kPa calibrated span that is a total band of ±0.075 kPa covering all three effects at once. General purpose pressure transmitters commonly sit in the 0.3 to 0.5 percent class.

Repeatability appears as its own line, always smaller than reference accuracy. Reference accuracy sets the total band; repeatability tells you how much of that band is random scatter rather than correctable shape error.

Temperature effect and stability are adders on top of reference accuracy. An instrument that reads within 0.075 percent on the bench can legitimately read several times that outdoors in winter, and the datasheet says so in the fine print.

Temperature sensors state accuracy as a tolerance class instead. An IEC 60751 Class A RTD sensor is allowed ±(0.15 + 0.002·|t|) °C, which is ±0.35 °C at 100 °C; Class B doubles the base term and widens to ±0.80 °C at the same temperature.

And note the basis of every percentage: percent of full scale, percent of span, and percent of reading are three different sizes of error band. That choice changes the real error more than most buyers expect, and we walk through the math in percent of full scale vs percent of reading.

Calibration removes bias

ISO 5725 gives the tidy version of all this. It splits accuracy into trueness, closeness of the average of many readings to the true value, and precision, the spread of those readings. The gap between the average and the truth is the bias.

Calibration attacks bias and only bias. A zero and span trim against a reference standard shifts the whole reading set onto the true value: Unit A’s +0.14 bar offset disappears, and its tight 0.007 bar standard deviation stays. Nothing in a calibration procedure can shrink random scatter; averaging multiple readings helps a data logger, but a control loop acts on single readings in real time.

That is the practical rule I give customers: bias is correctable, scatter is not. When you compare two instruments, look at repeatability first, because that is the floor of achievable performance after your first calibration. It also means a calibration is only as good as its reference; field practice is to use a reference at least four times more accurate than the device under test.

Resolution is different

Resolution is the smallest change an instrument can detect or display. It is a third property, not a synonym for either accuracy or precision, and it is the easiest one to be fooled by: a five-digit display does not make a five-digit instrument.

A digital indicator can show 6.1382 bar while the sensor behind it holds a 0.5 percent accuracy class, meaning the last two digits are decoration. Writing 6.1382 bar claims five significant figures when the accuracy class supports barely three; significant figures on a display are a resolution statement, not an accuracy claim.

The same trap appears at the control system input: a 12-bit analog input that spans exactly 4–20 mA resolves 4096 counts, steps of about 0.024 percent, so the DCS trend can move in fine steps while the transmitter feeding it is only accurate to twenty times that. We cover counts and loop resolution in the 4–20 mA signal conversion guide, and the loop itself in how a 4–20 mA current loop works.

Rule of thumb: resolution should be several times finer than the accuracy band, so the display never limits the sensor. The reverse arrangement, coarse accuracy behind fine resolution, only creates false confidence.

Choosing the right class

Accuracy costs money, and different services weight the two properties differently:

Service	What matters most	Typical class
Control loops	Repeatability. The loop holds a setpoint against its own reading, so a stable small bias barely matters; scatter feeds straight into valve movement.	0.2 to 0.5 percent
Custody transfer, billing	Accuracy with a traceable calibration certificate. Every percent of bias becomes a billing error.	0.1 to 0.2 percent, verified
Alarms, monitoring	Reliability and drift, not tight class. A high-level alarm at 90 percent does not need 0.075 percent accuracy.	0.5 to 1.5 percent

The class also depends on the measuring principle. A Coriolis mass flow meter is available down to the 0.1 percent class, while a vortex flow meter lives in the 1.0 to 1.5 percent of reading range no matter how much you spend. Asking a technology for a class it cannot physically reach is a common RFQ mistake, and the honest answer is a different technology or a compensation scheme.

FAQ

What is an example of accuracy but not precision?

Unit B in the worked example above: five readings of 5.96 to 6.05 bar against a true 6.00 bar. The mean is exactly right, so the instrument is accurate on average, but the scatter of ±0.05 bar means any single reading is not dependable. Wide scatter centered on the truth is accuracy without precision.

How to remember the difference between precision and accuracy?

ACcurate is Correct: accuracy means close to the true value. PRecise is Repeatable: precision means the readings agree with each other. If you can only remember one sentence, use this: calibration fixes accuracy; only a better instrument fixes precision.

How do you determine accuracy and precision?

Hold a known true value with a reference standard, take repeated readings, then compute two numbers. Accuracy is the difference between the mean of the readings and the reference value, usually expressed as percent error. Precision is the standard deviation of the readings, often quoted as a two or three sigma band. The reference should be at least four times more accurate than the instrument under test.

What is accuracy and precision in measurement?

Accuracy is closeness of a measurement to the true or accepted value; precision is closeness of repeated measurements to one another. ISO 5725 combines them: accuracy in the broad sense equals trueness, the systematic part, plus precision, the random part. An instrument spec covers both through its reference accuracy and repeatability lines.

Request a quote

Tell us the measurement, the operating point, and the accuracy class the application really needs, and we will match the instrument to it: pressure, differential pressure, flow, level, or temperature. If a cheaper class does the job, we will say so.

Written and technically reviewed by Wu Peng and the Instranova engineering team.