HRV is not one number. It is a family of statistics calculated from the gaps between your heartbeats, and every platform picks its own statistic, its own sampling window, and its own rules for throwing out bad beats.
That is why one person wearing five devices on one night gets five different numbers. In most cases none of them is broken. They are answering slightly different questions about the same heart. So how to measure heart rate variability comes down to two decisions: which statistic gets reported, and when the reading is taken.
The Verdict
What your device is actually measuring
The underlying measurement is the RR interval: the time from one heartbeat to the next, recorded in milliseconds. The name comes from the R wave, the tall spike in an ECG trace that marks a ventricular contraction. At a resting heart rate of 60 beats per minute, those intervals sit near 1,000 ms each and vary by a few tens of milliseconds from beat to beat.
Before any statistic is calculated, the software screens that list for beats that are not normal sinus beats. Extra beats, missed beats and noise get corrected or removed, and the cleaned list is called the NN series (normal-to-normal intervals). Every HRV metric worth reporting is computed from NN intervals, not raw ones.
So when your app shows "HRV 48 ms", the milliseconds do not describe your heart rate or the interval length. They describe how much those intervals moved, summarised by one particular formula. Which formula matters a great deal.
The metrics behind the number, and what each responds to
Four statistics account for almost everything you will see on a device or in a study. They take the same input and compress it differently, which is why they produce different values from identical heartbeat data.
| Metric | How it is calculated | What it mostly reflects | Recording length it needs |
|---|---|---|---|
| RMSSD | Take the gap between each pair of neighbouring NN intervals, square each gap, average the squares, then take the square root. | Fast beat-to-beat change, driven mainly by the vagus nerve. | Usable from about 60 seconds upward. That tolerance is why nearly every consumer device uses it. |
| SDNN | The standard deviation of every NN interval in the recording. | Total variability from both autonomic branches, including slow rhythms that take minutes to appear. | Strongly length-dependent. A 1-minute SDNN, a 5-minute SDNN and a 24-hour SDNN are three different measurements sharing one name. |
| pNN50 | The share of neighbouring interval pairs that differ by more than 50 ms, written as a percentage. | The same vagal activity RMSSD tracks, expressed as a percentage instead of milliseconds. | Needs a few minutes, and loses resolution in older adults whose intervals rarely differ by a full 50 ms. |
| LF and HF power | Break the interval series into frequency bands: LF at 0.04 to 0.15 Hz, HF at 0.15 to 0.40 Hz. | HF tracks breathing-linked vagal activity. LF mixes several influences and is not a clean sympathetic marker. | Needs a stable 5-minute recording at a controlled breathing rate before the numbers mean anything. |
RMSSD dominates consumer devices for one practical reason: it is stable over short windows. Because it works on the difference between neighbouring intervals, it captures fast vagal changes and is not waiting on slow rhythms to complete.
SDNN behaves in the opposite way, and this is the most common comparison error people make. A longer recording captures slower rhythms, so SDNN rises simply because the recording got longer. A 24-hour SDNN and a 5-minute SDNN are not comparable numbers, and neither is comparable to an SDNN taken from a 60-second sample.
The frequency-domain measures split the interval series into bands, following the definitions the European Society of Cardiology task force standardised for HRV reporting. Consumer apps mostly avoid them because they demand a still recording of at least five minutes at a controlled breathing rate, and because the old reading of LF/HF as a "sympathetic balance" ratio is now widely disputed.
What "ms" means on screen, and why some apps show a score instead
A value in milliseconds is the statistic itself, unmodified, which makes it the only HRV output you can compare against published ranges. An RMSSD of 45 ms means the typical difference between one heartbeat gap and the next was 45 thousandths of a second.
Several platforms transform that value before showing it. The most common transform is the natural log, written lnRMSSD. RMSSD is skewed, so a 10 ms change means something different at 25 ms than at 90 ms; taking the log makes changes proportional and easier to trend. Some apps then multiply lnRMSSD by a constant to spread it across a friendlier 0 to 100 range.
Other platforms skip the physiology entirely and score your reading against your own recent baseline. A recovery percentage or a 0 to 100 readiness figure is a ranking of today against your own past, not a measurement in any unit. Two apps both showing 70 are saying nothing about each other.
The practical rule is to find the millisecond value in your app if you want to place yourself against a published chart, and ignore the score for that purpose. Typical millisecond ranges by decade live on HRV by age.
When the reading is taken changes it more than what took it
Timing and posture move HRV further than any difference between two accurate devices. This is the single biggest reason two well-built platforms disagree, and it has nothing to do with sensor quality.
HRV is not flat across a night. Vagal tone generally climbs after sleep onset and runs highest in the middle of the night, then falls as morning approaches. A device averaging the whole night, a device sampling a slow-wave block at 3am, and a device you check at 7am are sampling three different points on that curve.
Posture does more still. Lying down produces the highest values; sitting is lower; standing withdraws vagal tone within seconds and can cut RMSSD to a fraction of your supine value. That change appears in under a minute, and it is larger than the change most people achieve from months of aerobic training.
Two consequences follow. First, an overnight-average device and a morning spot-check device will report different numbers for the same person on the same day, and both can be correct. Second, protocol consistency buys you more signal than upgrading hardware does, because an inconsistent posture or wake time introduces more variation than the sensor ever will.
Sensor tiers, from ECG down to wrist optical
Precision follows the sensor type, and the gap between tiers is larger than the gap between brands within a tier. Three tiers cover everything available to a consumer.
Single-lead ECG, meaning a chest strap or a clinical ECG, measures the heart's electrical activity and times the R wave directly. The R wave is a tall, narrow spike, so identifying its exact position is straightforward and timing resolution is high. This is the reference standard, and it is what clinical HRV norms were built from.
Finger or ear optical (PPG) sits one step down. An LED shines into tissue and a photodiode reads pulse-driven changes in blood volume. Fingers and ears have dense capillary beds close to the surface, so the waveform is strong when you are still.
Wrist optical is the loosest tier. The sensor sits over tissue with less blood flow, band fit varies, and the wrist moves constantly. Devices compensate with heavier filtering, which is why wrist platforms take their HRV reading during sleep rather than during the day.
Motion damages PPG far more than it damages a chest strap, and the reason is mechanical. A PPG sensor infers a pulse from a light signal, and movement changes the distance between sensor and skin and redistributes blood in the tissue, producing waveform bumps that look exactly like beats. An ECG strap reads voltage, and while motion adds noise, the R wave remains a distinctive spike the algorithm can still find. Cold matters for the same reason: vasoconstriction weakens the optical signal but leaves the electrical one untouched. For HRV specifically, a chest strap during a controlled five-minute reading is the consumer gold standard. What optical sensors produce is technically pulse rate variability, a closely related but not identical quantity, covered on what is HRV.
How the major platforms measure heart rate variability
Six platforms cover most of the consumer market, and no two of them made the same pair of choices. The table adds Fitbit, Garmin and Polar to the three already named, listed alphabetically by platform. Implementations change with firmware and app releases, so treat these as the general approach each vendor takes rather than a fixed specification.
| Platform | Metric reported | When it samples | What that makes it good for |
|---|---|---|---|
| Apple Watch | SDNN in milliseconds, written into the Health app | Short background samples taken across the day and night, plus during guided breathing sessions | Following your own long-run trend and exporting raw values. Daytime samples are noisy because posture and activity are uncontrolled. |
| Fitbit | An RMSSD-based nightly value in milliseconds | During sleep, averaged across the night | Steady night-to-night comparison. An all-night average is less sensitive to one restless hour than a short window is. |
| Garmin | An RMSSD-based overnight value, plus a status label against your own range | Across the sleep window, then compared to a rolling multi-week personal baseline | Seeing whether you have drifted off your own baseline without over-reading any single night. |
| Oura | RMSSD in milliseconds, as a nightly average and a plotted overnight curve | Repeatedly through the night, from finger PPG | Reading the shape of a night, including how late in the night HRV recovers after alcohol or a late meal. |
| Polar | An RMSSD-derived recovery value, plus raw beat-to-beat intervals from its chest straps | Early in the sleep period for recovery scoring, and on demand for orthostatic and strap-based tests | Controlled protocol readings, because the strap exports the interval series for you to analyse yourself. |
| Whoop | RMSSD in milliseconds | During a deep (slow-wave) sleep window late in the night | A physiologically matched window each night, which reduces the effect of an irregular bedtime on the number. |
Read down the middle column first. A platform reporting SDNN from minute-long samples and a platform reporting RMSSD from a slow-wave block are not producing comparable numbers, no matter how good either sensor is. Only after that does the sensor tier matter.
The right-hand column is the buying criterion most reviews skip. If you want a controlled reading you can repeat and export, a chest strap that hands you the raw interval series is the setup that supports it. If you want a number that appears without any effort from you, an overnight platform is the better fit, and the specific window it uses matters mainly for how stable it stays when your bedtime moves around.
Artefact correction and the improvement that is really an error
A single bad beat can inflate RMSSD dramatically, and this is the most under-reported failure mode in consumer HRV. The arithmetic explains why.
An ectopic beat arrives early, producing one unusually short interval, and is usually followed by a compensatory pause, producing one unusually long one. RMSSD squares the difference between neighbouring intervals, so those two abnormal gaps contribute enormously to the average before the square root is taken. A missed beat does the same damage from the other direction: the algorithm merges two real intervals into one of roughly double length.
The result is a reading that looks like the best recovery of your life. The signature is a jump well outside your normal range with no matching drop in resting heart rate and no change in how you feel. A genuine HRV rise and a resting heart rate fall move together, so a large HRV gain on its own is more likely an uncorrected artefact than a physiological event.
Kubios and similar analysis software make this step visible: you choose a correction level and the software reports what proportion of beats it altered. A useful working rule is that if more than a few percent of beats needed correction, discard the recording instead of interpreting it. Consumer platforms perform the same correction silently, with undocumented thresholds, which is one more reason two devices disagree on the same night.
There is a named exception where correction cannot rescue the reading. In atrial fibrillation, or with frequent ectopy, the intervals are genuinely irregular rather than artefactually so, and HRV software may report the arrhythmia as an outstanding score. An unusually high number in someone who feels unwell is a reason for an ECG, not for confidence.
How to measure heart rate variability at home, step by step
A repeatable protocol is worth more than a better sensor, because it removes the variables that swamp the signal. Six rules cover it.
- Same time every day. Within a few minutes of waking is the easiest slot to hold, and it sits before the day's stressors.
- Same posture. Pick supine or seated and never mix them. Standing readings are a separate test with separate values.
- Same conditions. After emptying your bladder, before caffeine, food, a shower, or checking your phone.
- Two to five minutes, first minute discarded. The opening period covers you settling down, and it drags the number around.
- Breathe normally, or fix a rate and keep it. Breathing rate directly changes HRV, so slow breathing during a reading inflates it against your own history.
- One device, one metric. Adding a second platform gives you two baselines rather than a cross-check.
Then compare correctly. Read a 7-day average against a 30-day average to judge the current week, and read the 30-day average against where it sat six months ago to judge a trend. Single readings are the noisiest layer available, and reacting to them is how most people end up frustrated with the metric. If your baseline is falling and you want the causes ranked, HRV explained covers what actually moves the number and why is my HRV low covers troubleshooting a specific low reading.
One rule outranks everything above it. Never compare your number to someone else's, and never compare it to your own number from a different device. A friend's screenshot, a published chart and your old wearable each used a different statistic, a different window, or a different sensor, so the gap between their number and yours carries no information at all. One device, one protocol, one baseline, tracked over months, is the whole of measuring HRV usefully.
When to talk to a physician
Stop tracking and get assessed if erratic HRV readings come with palpitations, dizziness, fainting, chest pain, or breathlessness that is out of proportion to your effort. Consumer HRV was never designed as a cardiac screen, and irregular intervals need an ECG to characterise properly.
Also raise it clinically if a device repeatedly flags an irregular rhythm, or if your HRV baseline has fallen over several weeks alongside a resting heart rate that will not settle. That combination has a short list of common explanations a clinician can test for, and none of them are visible in a wearable chart.
Frequently Asked Questions
Where do I find the HRV number in my app?
It is almost never on the watch face or the ring itself, which is why so many people conclude their device does not measure it. On Apple devices the value lives in the Health app on the phone, under the heart section, rather than on the watch. Fitbit, Garmin and Oura surface theirs inside the nightly sleep or readiness summary rather than the main dashboard, usually one tap below the score. If you want the millisecond series rather than a chart, look for the data export option in account settings, since that is the only route to numbers you can analyse yourself.
Does HRV change across the menstrual cycle?
Yes, enough to move a weekly average without anything else changing. Cycle phase is one of the physiological rhythms that shifts autonomic tone, so a seven-day average compared against a thirty-day average can drift for reasons that have nothing to do with training, sleep or stress. The measurement consequence is what matters: compare like with like by reading the same phase against the same phase in previous cycles, rather than treating a mid-cycle dip as a recovery problem. Devices that log cycle information alongside HRV make that comparison easier.
Can HRV be measured during pregnancy?
The same sensors work normally in pregnancy, but the baseline they are judged against moves. Resting heart rate rises across gestation and blood volume increases, so a personal baseline built before conception stops being the right reference point within weeks. Read the trend over a short recent window rather than against your pre-pregnancy history, and expect readiness-style scores to look pessimistic while the algorithm catches up. A consumer wearable is not a maternal or fetal monitor, so anything that concerns you belongs with your midwife or obstetrician rather than with the app.
Is an HRV reading taken while ill still valid?
The measurement is valid; what changes is the meaning, not the accuracy. Fever, dehydration and an elevated resting heart rate all push HRV down, and the sensor is reporting that correctly rather than malfunctioning. The practical issue is your baseline: a week of illness readings drags a rolling thirty-day average down, so the fortnight after you recover can look like a deficit that has already resolved. Note illness days, keep taking readings if it is easy, and judge your recovery by the direction of travel rather than by the distance to a baseline the illness moved.
Is an HRV reading after a night shift comparable to a normal one?
No, and the reason is timing rather than the device. HRV follows a daily rhythm, so a reading after daytime sleep is taken at a different point on that curve than your usual post-overnight reading, and the two are not interchangeable even from the same sensor on the same person. Shift workers get more from two separate baselines, one for day-sleep readings and one for night-sleep readings, than from a single average that mixes them. If you only keep one, tag shift days so you can see which points came from them.
Can I use the ECG app on a watch to measure HRV?
Not usefully, because those recordings are too short and the interval data is generally not released. A consumer single-lead ECG captures a fixed window of roughly half a minute, and while the electrical trace is clean, a stable HRV statistic needs a longer settled recording than that. Most ECG apps also export a waveform or a PDF rather than the beat-to-beat interval series, which is the input any HRV calculation needs. A chest strap that hands you the intervals is the practical route to an ECG-quality HRV reading.
Is Fitbit HRV accurate?
For overnight trend tracking, wrist optical sensors including Fitbit agree reasonably well with ECG, because you are still and the signal is clean. Accuracy falls away in two situations: during movement, when the optical waveform is corrupted by the sensor shifting against the skin, and in cold conditions, when reduced blood flow to the skin weakens the pulse signal. The practical limit is not the brand. It is that any wrist device measures pulse timing rather than the heart's electrical activity, so treat the nightly trend as the signal and any single night as approximate.
Can I measure HRV with a phone camera or an app alone?
Yes, and the accuracy depends entirely on how still you can hold your finger. A phone camera app works as a PPG sensor: the flash illuminates your fingertip and the camera reads pulse-driven colour changes. Under a controlled, motionless, well-lit reading the intervals can be usable, but the frame rate limits timing resolution and any pressure change on the lens corrupts the waveform. Camera-based readings are the least precise consumer option, and they are worth using only if you keep the protocol identical every single time.
Related
- HRV explained: what actually raises and lowers the number
- HRV by age: typical millisecond ranges by decade
- Why is my HRV low: troubleshooting a specific reading
- Oura vs Whoop: two overnight platforms on one rubric
- Wearables compared: the full device rubric