Accuracy and repeatability describe two different things an instrument can do well or badly, and confusing them leads to bad instrument choices. Accuracy is how close a reading is to the true value; repeatability is how consistently the instrument gives the same reading for the same input. A sensor can be extremely repeatable while being consistently wrong, which is exactly why the distinction matters when you pick instruments for control loops versus custody transfer.
Accuracy vs Repeatability in one line: Accuracy is how close a measurement is to the true value, while repeatability is how consistently an instrument returns the same reading for the same input under the same conditions. An instrument can be highly repeatable yet inaccurate if it has a stable bias, which is why the two specs must be judged separately.
Picture a target. Accuracy is whether your shots land on the bullseye; repeatability is whether they land in a tight cluster, wherever that cluster is. Shots can be tightly grouped but off to one side, which is high repeatability with poor accuracy, or scattered around the center, which is decent average accuracy with poor repeatability. A good instrument does both, but the two properties are independent and are specified separately.
A vendor's accuracy figure is usually a total error band rolled up from several contributors. Hysteresis is the difference in reading depending on whether you approached the value from above or below, so a transmitter reads slightly differently going up in pressure than coming back down. Linearity, or nonlinearity, is how much the response deviates from a straight line across the span. These, along with repeatability itself, combine into the accuracy or total-error-band number on the datasheet.
Repeatability is often the best of these numbers because it excludes bias, hysteresis over the full span, and drift. That is not a trick; it reflects a real distinction. An instrument that returns to the same value every time for the same input is repeatable even if that value is offset from the truth, and for some jobs repeatability is exactly what you need.
The key insight is that a stable, repeatable bias is invisible to the instrument itself. If a pressure transmitter reads three psi high every single time, it is perfectly repeatable and perfectly wrong. Repeatability tells you the reading is trustworthy relative to itself; it says nothing about whether it is trustworthy relative to the true value. Only calibration against a traceable reference reveals and corrects the bias.
This is why calibration matters even for a beautifully repeatable instrument. Repeatability guarantees that once you know the offset, it will stay put and can be corrected. An instrument that is repeatable but uncalibrated is like a clock that runs perfectly but is set to the wrong time: reliable, consistent, and useless for telling the actual time until you set it correctly.
Drift is the enemy of both. Over months an instrument can develop a growing bias, degrading accuracy, and if it also becomes erratic, its repeatability suffers too. Regular calibration re-anchors accuracy, while a device that will not hold repeatable readings between calibrations is failing at the more fundamental of the two properties and usually needs replacement.
A closed control loop mostly cares about repeatability. If a level transmitter reads consistently, the controller can hold the level at a stable, repeatable point even if that reading carries a small bias, because the operator simply sets the target to the value that produces the desired result. A repeatable but slightly biased sensor gives smooth, stable control. An erratic sensor, even a nominally accurate one, makes the loop hunt and oscillate.
Custody transfer is the opposite. When a measurement determines who owes whom money, a stable bias is not harmless, it is a systematic billing error in someone's favor. Fiscal and allocation measurement therefore demands genuine accuracy traceable to a reference standard, with calibration and often proving to prove the absolute number, not merely its consistency. This is why fiscal meters are held to far tighter, independently verified accuracy than a typical process instrument.
On a cloud SCADA platform like Merobix, both properties show up in the trends but in different ways. A jumpy, non-repeatable signal is visible immediately as noise and is usually a fault or a poor installation. A stable bias is invisible in the trend alone and only surfaces when the reading is compared against a known reference or a redundant measurement, which is why scheduled calibration, not just live monitoring, remains essential for the numbers that have to be absolutely right.
Accuracy is how close a reading is to the true value, while precision, in instrument terms usually called repeatability, is how consistently the instrument returns the same reading for the same input. They are independent: a sensor can be precise but inaccurate if it has a stable offset, or accurate on average but imprecise if its readings scatter. Both are specified separately on a datasheet.
Yes, and it is common. An instrument with a stable bias returns the same reading every time for a given input, which is perfectly repeatable, yet that reading is consistently offset from the true value, which makes it inaccurate. The bias is invisible until the instrument is calibrated against a traceable reference, which is why calibration is required even for very repeatable devices.
A control loop holds a process at a setpoint, and it can do that with a repeatable but slightly biased sensor because the operator sets the target to whatever reading produces the desired result. Consistency keeps the loop stable, while a small fixed offset just shifts the setpoint. Custody transfer is different: there a stable bias becomes a systematic billing error, so it demands genuine traceable accuracy.
Merobix reads your field devices into a cloud SCADA - the real thing behind these terms, live in days from any browser.