CGM accuracy: what MARD means, and what it hides
Updated 2026-08-07
Short answer
What does MARD mean on a glucose sensor?
MARD — mean absolute relative difference — is the average percentage gap between what a sensor reports and what a laboratory blood test says at the same moment. A MARD of 9% means readings are off by about 9% on average, so a true 7.0 mmol/L might display anywhere around 6.4 to 7.6. Lower is better, but because it is an average it says nothing about the worst readings.
How to read the evidence labels
Every claim on this page carries a label saying how much weight it can hold. The label describes the strength of the finding, not the prestige of the journal it appeared in.
- Strong evidenceA large randomised trial, a long cohort study, or agreement between several of them.
- Emerging evidenceReal, but thin — small, short, observational, or not yet replicated.
- ContestedCompetent researchers actively disagree. We are not picking a side for you.
What the number is
A sensor is worn while blood is drawn repeatedly and measured on a laboratory analyser. Each paired comparison gives a percentage difference; MARD is the mean of those percentages. Modern sensors sit under 10%. Twenty years ago they were over 20%.
Being an average, it is silent about the tail. A device with a good MARD can still produce occasional readings that are badly wrong, and it is those readings, not the average, that cause a wrong decision.
What the evidence says
Current continuous glucose monitors report MARD values around or below 10%, a substantial improvement over earlier generations of the technology.1
Strong evidence
Why two MARD figures are often not comparable
MARD depends on how the study was run: which reference method was used, how much of the sampling happened while glucose was moving quickly, how many days of wear were included, and which population was recruited. Change those and the same sensor produces a different number.
This is why we record where each figure came from rather than only what it is. An independent head-to-head study that put two flagship sensors on the same participants reported worse figures for both than each manufacturer's own pivotal trial did — which is what you would expect, and exactly why the source belongs beside the number.
What the evidence says
An independent comparison of two widely used sensors reported point accuracy materially worse than the manufacturers' own pivotal trials for both devices.2
Emerging evidence
How we tier accuracy on this site
Every accuracy figure on a device page carries one of three labels. An independent study is one whose authors are not employed by the manufacturer. A manufacturer study is a properly conducted trial that lists company employees among its authors — normal for registrational research, and still worth knowing. A manufacturer-stated figure has no published study behind it at all, only a claim on a product page.
We do not borrow a figure from a sibling product. A Libre 3 number is not a Libre 3 Plus number, and a G7 number is not a Stelo number, even where the hardware is related. Where no figure exists, the page says so.
Common questions
- Is a 9% MARD good?
- It is typical of current sensors and much better than the technology managed a decade ago. Whether it is good enough depends on the decision you are making: it is adequate for spotting trends, and it is why insulin dosing decisions are still confirmed with a fingerstick when a reading conflicts with symptoms.
- Why do some devices show no accuracy figure here?
- Because we could not find one published for that specific product. We do not substitute a related product's figure, and we do not estimate. A blank means unknown, not zero.
- Should I pick the sensor with the lowest MARD?
- Only if the figures came from comparable studies, which they often did not. Reading interval, alarms, wear length and whether the device is sold where you live usually matter more to the decision than a percentage point of MARD.
References
- 1.Almurashi AM, Rodriguez E, Garg SK (2023). Emerging Diabetes Technologies: Continuous Glucose Monitors/Artificial Pancreases. Journal of the Indian Institute of Science. 10.1007/s41745-022-00348-3
- 2.Hanson K, Kipnes M, Tran H (2024). Comparison of Point Accuracy Between Two Widely Used Continuous Glucose Monitoring Systems. Journal of Diabetes Science and Technology. 10.1177/19322968231225676