The short version
- A 2024 biological variation study sampling 20 healthy adults six times a day for five days found the reference change value for salivary cortisol is 96% to 245%, meaning a repeat result must roughly double or triple before the change is real.
- Within-person biological variation in salivary cortisol ranged from 29.3% to 56.5% across the day in that study, measured by mass spectrometry with every sample run in duplicate.
- A 2013 study found that delaying the first waking saliva sample by 5 to 15 minutes significantly inflated the estimated morning cortisol rise, and that self-reported waking times were a median of four minutes later than actigraph-measured waking.
- Cortisol does not begin to rise significantly until about 10 minutes after waking, which is why a sample taken slightly late produces a systematically wrong estimate.
- When one Swedish group split the same saliva samples across six assays, the upper reference limit for a late-night sample was 3.4 to 3.9 nmol/L on three mass spectrometry methods but 5.8, 4.3 and 21.6 nmol/L on three immunoassays.
Here is the number that should decide this for you, and almost nobody selling a kit will print it.
In 2024 a group at the University of Verona took 20 healthy volunteers and had them collect saliva immediately on waking and then at 08:00, 12:00, 15:00, 19:00 and 23:00, every day, for five consecutive days. Every sample was analysed in duplicate by mass spectrometry — the reference method, not a consumer assay — and the variation was split statistically into its analytical and biological components.
Within a single person, the biological variation in salivary cortisol ranged from 29.3% to 56.5% depending on the time of day. From that they calculated the reference change value: how different two results have to be before the difference is real rather than noise. For cortisol it came out at 96% to 245%.
In plain terms, if you test today and test again next month, the second number has to be roughly double to triple the first before anyone can honestly say something changed. A report telling you that you are 20% or 30% above some optimal band is describing an amount of movement this hormone produces on its own, on a normal Tuesday, in a healthy person who did nothing at all.
Five minutes decides your morning result
Most kits ask for a waking sample, because the morning rise is the interesting part of the curve. Smyth and colleagues, in Psychoneuroendocrinology in 2013, tested how precise that has to be. Fifty healthy women collected saliva at 0, 15, 30 and 45 minutes after waking on four days, with both their waking and their sampling times electronically monitored — actigraphy on the wrist, and caps that record when the tube is opened.
Self-reported waking was later than the actigraph estimate by a median of four minutes. On days when the first sample was delayed by 5 to 15 minutes, the estimated size of the morning rise came out significantly larger than on days when the delay was under 5 minutes, and the peak appeared to arrive earlier. A second, intensive study sampling every 5 minutes found cortisol did not begin to rise significantly until 10 minutes after waking.
So a delay everyone previously assumed was tolerable changes the answer, in a predictable direction, and you cannot reliably report your own waking time to within five minutes. Which is why the researchers use wrist actigraphs and electronic caps and a kit uses your memory of what time you got up.
And the threshold depends on which machine reads it
A Swedish multi-centre group in 2023 collected saliva from 155 people in a reference population and 22 patients with confirmed cortisol excess, then split each sample and ran the aliquots through three mass spectrometry methods and three immunoassays. Same saliva, six methods.
The upper limit of normal for a late-night sample came out at 3.4 to 3.9 nmol/L across the three mass spectrometry methods — reassuringly tight. Across the immunoassays it was 5.8, 4.3 and 21.6 nmol/L. The highest and lowest differ by a factor of five. A single result of, say, 6 nmol/L is above the reference limit on one platform and unremarkable on another, and the number on your report does not tell you which one you got.
What a single reading cannot tell you
Put those three findings together and a one-off consumer cortisol result cannot tell you whether your level has changed, cannot be timed accurately enough at the one moment that matters most, and cannot be compared to a threshold without knowing the assay. Three independent reasons, each sufficient on its own.
What it can do is produce a colour-coded band that feels like information. That band will either reassure you falsely or worry you about a hormone doing its job, and in both cases you will probably buy something.
What we'd actually tell you
Don't buy the kit. If you want to track how a stressful stretch is actually going, track things that do not have a 245% reference change value: hours slept, whether you woke rested, resting heart rate, how many days this week you did the thing that helps. A fortnight of that will tell you more than a saliva tube, and slow breathing is the one intervention in this whole category with pooled randomised data and no purchase attached.
If you have symptoms that genuinely worry you — unexplained weight or skin changes, persistent weakness, blood pressure that has moved, anything progressing — that is a doctor, and the diagnostic pathway they use is timed, confirmed and repeated for exactly the reasons above. And if the honest description of your situation is burnout rather than illness, no measurement is going to tell you something you do not already know.
Good questions
Can a saliva cortisol kit tell me if I am stressed?
No. It can tell you the concentration in one tube of spit at one moment, measured on an assay whose reference limit may differ fivefold from another lab's. Because a repeat result has to change by 96 to 245 percent to count as real, a single reading cannot establish that your level is high, low, or moving in any direction.
What about a four-point cortisol curve — is that better?
It is better than one sample and still not diagnostic at home. The curve's most informative section is the first 45 minutes after waking, and that section is the hardest to collect correctly. Researchers electronically monitor both waking time and tube-opening time because self-report is not accurate enough, and a five-minute error changes the shape of the curve.
Why do two labs give me different cortisol results?
Because they are probably running different assays. In a study that split identical saliva samples six ways, the upper reference limit for a late-night sample ranged from 4.3 to 21.6 nmol/L across three immunoassays. Your result is not comparable across providers, and neither is the normal range printed beside it.
I already bought a test and it came back high. What now?
Do not act on it alone, and do not buy the protocol that was recommended alongside it. If you have symptoms that genuinely concern you — unexplained weight or skin changes, persistent weakness, new blood pressure problems, anything progressing — take those symptoms to a doctor. They will use timed, repeated, confirmed tests, which is the part a home kit cannot replicate.
What should I track instead?
Things with less inherent noise than a hormone that varies by up to 56% within the same person. Hours slept, whether you woke rested, resting heart rate from a wearable, and a simple daily note on how the day went. Two weeks of that is more actionable than any single saliva result, and it costs nothing.
Sources
- Danese E, Padoan A, Negrini D, et al. Diurnal and day-to-day biological variation of salivary cortisol and cortisone. Clin Chem Lab Med, 2024. View study
- Smyth N, Clow A, Thorn L, Hucklebridge F, Evans P. Delays of 5-15 min between awakening and the start of saliva sampling matter in assessment of the cortisol awakening response. Psychoneuroendocrinology, 2013. View study
- Bäcklund N, Brattsand G, Lundstedt S, et al. Salivary cortisol and cortisone in diagnosis of Cushing's syndrome - a comparison of six different analytical methods. Clin Chem Lab Med, 2023. View study
These statements have not been evaluated by the Food and Drug Administration. This product is not intended to diagnose, treat, cure, or prevent any disease.
