Which home blood pressure monitor is accurate, and does cuff size matter more than the device?

Nothing on this page came off our own bench

This desk does not own a blood-pressure rig. There is no clause for blood pressure on our methods page, because writing one would mean acquiring a reference — trained, blinded, double-observer auscultation against a calibrated standard, on a pre-declared number of subjects — and we have not. So this report runs under clause M-09: every figure below was produced by somebody else, and it says who, in the same row as the number.

What we did on 2026-08-10 was read the three public registries of validated devices, pull the published validation studies for models the makers named here actually sell, and read the makers’ own published cuff dimensions. What we did not do is put a monitor on a bench. When we can, we will, and this page will be superseded rather than quietly edited.

This is health information, not medical advice. Nothing here diagnoses anything. Home readings are worth taking to an appointment. They are not worth acting on alone, and no reading from any device on this page is a reason to start, stop or change a medication.

The error is not in the device

The interesting thing about home blood pressure is that the regulated part is the small part. A device that passes the AAMI/ESH/ISO Universal Standard is allowed a mean error of up to 5.0 mmHg. Two randomised crossover trials out of Johns Hopkins have since put numbers on the unregulated part — the cuff you chose and the way you were sitting — and those numbers are larger.

Sources: Ishigami et al., Effects of Cuff Size on the Accuracy of Blood Pressure Readings: The Cuff(SZ) Randomized Crossover Trial, JAMA Internal Medicine, 2023 (195 adults recruited from the community, Baltimore); Liu et al., Arm Position and Blood Pressure Readings: The ARMS Crossover Randomized Clinical Trial, JAMA Internal Medicine, 2024 (133 adults, Baltimore). Neither figure is ours. Both are read against the guideline-recommended position: seated, back supported, feet flat, arm supported on a desk with the cuff at heart level.

Read the first and the last rows together. The single largest device-side error the standard will tolerate is 5.0 mmHg. Putting a standard cuff on an arm that needs an extra-large one produces nearly four times that, in the same direction, on every reading. The ARMS numbers stack on top: resting the hand in the lap added 3.9 mmHg systolic and 4.0 mmHg diastolic; letting the arm hang at the side added 6.5 systolic and 4.4 diastolic. Both are ordinary things that people do at a kitchen table.

That is the whole argument of this page. Choosing between validated monitors is a small optimisation. Choosing the cuff, and sitting still with your arm on a table, is the large one.

Cuff size, from the arm you actually have

Measure the bare upper arm at its midpoint, halfway between the shoulder and the elbow, arm relaxed at your side. The circumference decides the cuff. The American Heart Association’s 2019 scientific statement gives the mapping.

The underlying rule, which is worth knowing because it explains the boundaries: the bladder inside the cuff should be 75% to 100% of the arm’s circumference in length and 37% to 50% of it in width. A bladder that is too short cannot occlude the artery without over-inflating, and the reading comes out high. Too long and it comes out low.

Two practical consequences. First, “universal” or “wide-range” cuffs are a compromise across that whole span, not a fifth size — they are useful, and they are still a range with ends. Second, if your measurement sits within a centimetre or two of a boundary, measure again on another day before you buy.

What the makers publish

Four makers, the specific models with published validations we could find, and the cuff ranges each one publishes. Checked 2026-08-10.

Omron publishes its cuff dimensions in inches — 7–9, 9–13, 13–17 and 9–17. The centimetre figures in that table are ours, converted at 2.54 cm per inch and rounded to the nearest centimetre. The inch figures are the ones printed on the product.

One column in that table does more work than the others: cuff sizes the maker publishes. Withings and Beurer both solve the problem with a single 22–42 cm cuff, which covers the small-adult, adult and most of the large-adult band in one strap. That is genuinely convenient, and for the large majority of arms it is correct. Omron instead sells four cuffs and expects you to pick. That is more friction at purchase and it is the reason it is our pick: the failure mode this page is about is buying a monitor whose cuff does not fit your arm, and a maker who sells you a small cuff and a large cuff separately has made that failure harder.

The 42 cm ceiling

Look back at the first table. The largest documented error, by a distance, is the extra-large case: +19.5 mmHg. Now look at the cuff ranges. Three of the four makers here stop at 42 or 43 cm. The AHA table has a band above that — 45 to 52 cm — and mainstream home monitors mostly do not serve it.

So the group with the largest known measurement error is also the group with the fewest validated options. That is not a criticism of any one maker; it is a description of the category. A&D Medical publishes an extra-large cuff at 42–60 cm, which is why it is on this page at all, but the validation it cites is to ANSI/AAMI SP-10, a protocol that predates the 2018 Universal Standard, and we did not find that model on the registries we checked. We are reporting the gap rather than filling it with a recommendation we cannot support.

If your measured arm circumference is above 42 cm, the useful move is not to buy a monitor and hope. It is to take the tape-measure number to a clinician or pharmacist and ask what they use for arms in that band, because a reading taken with the wrong cuff is worse than no home reading at all — it is a wrong number that looks like data.

Wrist cuffs

Wrist monitors are not fraudulent. Some wrist-cuff devices have passed the same protocols as upper-arm devices, and the registries list wrist-cuff and wrist-watch-cuff categories separately. The case against them at home is not that the hardware cannot measure. It is hydrostatics.

Pressure in a fluid column rises with depth at about 0.8 mmHg per centimetre for blood. A wrist resting in the lap sits roughly 20 cm below the heart, which adds on the order of 15 mmHg to the systolic reading before the sensor has done anything at all. An upper-arm cuff, on a seated person with the arm on a table, is close to heart level almost by default. A wrist has to be deliberately held there, on every single reading, forever.

The market compounds it. In the survey of the Australian online marketplace published in Hypertension in 2020, 972 unique devices were on sale across 59 online businesses: of 278 upper-arm cuff devices 18.3% were validated, of 162 wrist-cuff devices 8.0% were, and of 532 cuffless wrist-band wearables, none were. Zero out of 532.

That last figure is the one to carry. The 2025 AHA/ACC hypertension guideline advises against relying on cuffless devices, including smartwatches, until they demonstrate adequate accuracy and reliability. The same guideline treats home monitoring itself as central to diagnosis, which is exactly why the device doing it has to be a cuff on an upper arm.

Buy a wrist monitor when an upper-arm cuff genuinely will not work — arm circumference outside every available cuff, pain, lymphoedema, a wound. Then check the specific model on a registry first, and support the wrist at heart level for every reading.

Maker by maker

Omron Healthcare — our pick, on cuff range rather than on electronics. Models validated under the current Universal Standard exist (the J760 study is a 2025 publication), and the separately sold small, standard, large and wide-range cuffs mean a household with two very different arms can be measured correctly with one base unit. If you buy one thing off this page, buy an Omron upper-arm model and the cuff that matches your measured circumference, not the cuff in the box.

Withings — the best-documented single-cuff option here. BPM Connect and BPM Core both have peer-reviewed validations published under the Universal Standard, and BPM Connect’s validation explicitly covers the 22–42 cm general population, which is a more useful sentence than most accuracy marketing. The single cuff is the constraint: correct for most arms, unavailable for the band above 42 cm.

Beurer — a credible validated option with an older paper trail. The BM 28’s published validation used ESH-IP 2010, a protocol STRIDE BP still accepts but which the 2018 Universal Standard replaced. That is not a claim the device is inaccurate; it is a claim that the evidence behind it is older than the evidence behind the other two. We could not confirm a current Beurer model validated under the 2018 standard, and we are recording that as an open question rather than as a fault.

A&D Medical — on this page for one reason: it publishes a cuff that goes to 60 cm. Everything else about it we left uncertain, deliberately, because we did not confirm the model on a registry and its cited protocol is an old one.

What would change our mind

Three things, stated in advance so that they count.

1. A validated model with a cuff above 45 cm. If any maker publishes a validation under ISO 81060-2:2018 covering arms in the 45–52 cm band, that model becomes the most important device in the category, because it serves the population with the largest documented error. It would move ahead of the pick immediately.
2. A current Beurer validation under the 2018 standard. That would close the only real gap between Beurer and Withings on this page.
3. Our own rig. If this desk acquires a blood-pressure reference — two trained observers, blinded, against a calibrated standard, with a pre-declared sample size — we will run these devices ourselves and publish the residuals. Until then, everything above is a reading of other people’s evidence and it is labelled as such. Cross-lab replication is the standard we apply to other people’s results, and on this category we have not yet produced a first result of our own to replicate.

Anything we get wrong here goes in the corrections log with a date, not into a silent edit.