What Is a Normal CGM Range in mmol/L?
A continuous glucose monitor worn by an adult without diabetes typically reads between 3.9 and 7.8 mmol/L for roughly 95 to 97 percent of the day. The mean sits near 5.0 to 5.5 mmol/L, the standard deviation near 0.6 to 1.1 mmol/L, and the coefficient of variation usually under 20 percent. Post-meal peaks in that population commonly land between 6.5 and 8.5 mmol/L and return to baseline within two hours. Fasting and overnight values cluster between 4.2 and 5.6 mmol/L, with brief sensor readings in the high 3s that are usually artifact rather than true hypoglycemia. Sustained readings above 11.1 mmol/L, or repeated verified readings below 3.0 mmol/L, are the two patterns that should go to a clinician rather than into a spreadsheet.
The reference numbers in mmol/L
To convert, divide mg/dL by 18.0182. The clinically anchored thresholds are cleaner in mg/dL and slightly awkward in mmol/L. That is why you see 3.9, 7.8, 10.0 and 13.9 everywhere.
| Threshold | mmol/L | mg/dL | What it marks |
|---|---|---|---|
| Level 2 hypoglycemia | < 3.0 | < 54 | Clinically significant low |
| Level 1 hypoglycemia | 3.0–3.8 | 54–69 | Alert low |
| Tight range floor | 3.9 | 70 | Bottom of the standard target range |
| Post-meal ceiling, non-diabetic | 7.8 | 140 | Upper bound of normal excursions |
| Target range ceiling, diabetes | 10.0 | 180 | Top of the 3.9–10.0 time-in-range band |
| Level 2 hyperglycemia | > 13.9 | > 250 | Sustained high |
Two range definitions matter and they are not interchangeable. The 3.9 to 10.0 mmol/L band is the consensus target for people with diabetes, where the goal is at least 70 percent of the day in range. The 3.9 to 7.8 mmol/L band is the one that describes healthy physiology. If you are wearing a sensor out of curiosity and you score 100 percent time in 3.9 to 10.0, you have learned almost nothing. Compute your time in 3.9 to 7.8 instead, and look at where the remaining minutes fall.
There is also no evidence that pushing an already normal profile flatter improves anything. A glucose curve is a readout of meals, sleep, illness and exercise. It is not a score to minimize.
Sensor error sits underneath every one of those numbers
A CGM measures glucose in interstitial fluid with an enzymatic electrode, applies a factory calibration, and smooths the signal. It does not measure blood glucose. Two consequences follow. There is a physiological lag of roughly 5 to 10 minutes during rapid change, so your true peak is earlier and higher than your curve shows. And there is a device error, reported as mean absolute relative difference (MARD), which in healthy adults on current sensors runs near 9 to 10 percent against a venous reference.
A 9 percent MARD at 5.5 mmol/L is about 0.5 mmol/L. That is larger than most of the differences people try to interpret. It also means that the boundary between “normal” and “slightly high” on a single reading is inside the noise band of the device. Trends and distributions are interpretable. Individual points near a threshold are not.
Error is not uniform across contexts either. Accuracy degrades in critical illness, where perfusion and edema change the tissue the sensor sits in. Studies in intensive care populations report worse agreement than in ambulatory adults.1 Similar work after liver transplantation found real-time CGM feasible in early postoperative care but required careful assessment of agreement before the data could guide anything.2 In chronic kidney disease, uremia and dialysis-related glucose shifts complicate both sensor accuracy and the usual relationship between average glucose and HbA1c.3 The sensor also behaves differently depending on whether it streams continuously or only when scanned. Real-time CGM outperformed intermittently scanned flash monitoring for glucose control in a randomized trial in type 1 diabetes.4
Three artifacts to recognize in your own data before you interpret anything:
- Compression lows. A sudden drop to 3.0 to 3.6 mmol/L that recovers within 20 to 40 minutes, almost always overnight, usually means you slept on the sensor. Check whether the dip coincides with a consistent sleep position and whether it recovers without you eating.
- First-day drift. Many sensors read low or noisy for the first 12 to 24 hours after insertion. We drop day one entirely from any summary statistic.
- Gaps. Signal loss produces missing 5-minute epochs. If you compute time in range over a denominator of available readings rather than elapsed time, gaps silently bias the result.
Computing your own metrics from the raw export
Dexcom Clarity exports a CSV with an EGV (mmol/L) or Glucose Value (mg/dL) column at 5-minute resolution. Abbott’s Libre export gives a Historic Glucose column at 15 minutes plus separate scan values. Nightscout gives you entries documents with an sgv field in mg/dL and an epoch date in milliseconds. All three become the same object: a timestamped series on a regular grid.
import pandas as pd, numpy as np
df = pd.read_csv("clarity_export.csv", parse_dates=["Timestamp (YYYY-MM-DDThh:mm:ss)"])
df = df.rename(columns={"Timestamp (YYYY-MM-DDThh:mm:ss)": "ts",
"Glucose Value (mg/dL)": "mgdl"})
df = df.dropna(subset=["mgdl"])
df["mmol"] = pd.to_numeric(df["mgdl"], errors="coerce") / 18.0182
# 5-minute grid, no interpolation across gaps longer than 15 min
s = df.set_index("ts")["mmol"].resample("5min").mean()
s = s.interpolate(limit=3)
# drop the first 24 h of each sensor session before summarising
start = s.first_valid_index()
s = s[s.index > start + pd.Timedelta("24h")]
expected = len(s)
valid = s.notna().sum()
print(f"coverage {valid/expected:.1%}") # want >= 85%
def pct(series, lo=-np.inf, hi=np.inf):
return ((series >= lo) & (series < hi)).sum() / series.notna().sum() * 100
print(f"mean {s.mean():.2f} mmol/L")
print(f"SD {s.std():.2f} mmol/L")
print(f"CV {s.std()/s.mean()*100:.1f} %")
print(f"TIR 3.9-7.8 {pct(s, 3.9, 7.8):.1f} %")
print(f"TIR 3.9-10.0 {pct(s, 3.9, 10.0):.1f} %")
print(f"below 3.9 {pct(s, hi=3.9):.2f} %")
print(f"above 7.8 {pct(s, lo=7.8):.2f} %")
print(f"GMI {12.71 + 4.70587*s.mean():.2f} %")
Interpret those outputs in this order. Coverage first, because a summary over 60 percent of the wear period is not a summary. Then the coefficient of variation, which tells you whether the mean is even meaningful: above roughly 36 percent the distribution is wide enough that the average hides the behavior. Then time in 3.9 to 7.8, then the tails.
GMI (glucose management indicator) is the estimated HbA1c implied by your mean CGM glucose. It disagrees with a measured HbA1c by 0.3 percentage points or more in a large fraction of people, because red cell lifespan and glycation rate vary between individuals. A stable GMI-minus-HbA1c offset is itself a useful personal parameter, and standard CGM reporting builds on exactly these summary metrics.5
Two further analyses are worth the effort. Segment by time of day and look at the 00:00 to 06:00 window separately, since overnight mean and variability are mostly free of meal input. And align windows to meal timestamps to get incremental area under the curve over 120 minutes, which compares two breakfasts far better than peak height does.
When a number needs a clinician and more analysis will not help
Take the following to a physician rather than to your notebook. Any fasting reading confirmed by fingerstick above 7.0 mmol/L, sustained readings above 11.1 mmol/L, or verified readings below 3.0 mmol/L in someone not taking glucose-lowering medication. Also take any symptoms of low glucose such as sweating, confusion or palpitations at any sensor value. Confirm with a meter before acting on a low, always.
Context also changes what “normal” means. Glycemic targets in inherited metabolic disease are set by the disease, and prospective CGM work in glycogen storage disease type Ia was needed just to define them.6 Endurance exercise produces excursions in both directions that would look abnormal at rest, as observational CGM data in para cyclists shows.7 Interpretation against a personal baseline beats interpretation against a population band.
Questions people also ask
What is an alarming blood sugar level in mmol/L? Below 3.0 mmol/L confirmed by fingerstick, or above 11.1 mmol/L sustained over hours, are the two patterns that warrant medical attention rather than self-analysis. A single sensor reading near either threshold is often artifact, given typical device error near 9 to 10 percent.
What is a good average glucose level in mmol/L? In adults without diabetes, mean CGM glucose usually falls between 5.0 and 5.5 mmol/L over two weeks of wear. The variability metrics matter as much as the mean, so a coefficient of variation under about 25 percent alongside that mean describes a stable profile.
What is a bad score for a glucose test? Diagnostic thresholds come from laboratory plasma tests, not CGM: fasting plasma glucose of 7.0 mmol/L or higher, or a two-hour value of 11.1 mmol/L or higher on an oral glucose tolerance test. CGM data can suggest that a lab test is worth ordering, and a clinician interprets the result.
What is a worrisome glucose level on a sensor? Repeated post-meal peaks above 10.0 mmol/L, a two-week mean above 6.5 mmol/L, or frequent verified lows in someone not on glucose-lowering medication. Each is a reason to get a plasma glucose and HbA1c measured rather than a conclusion in itself.
Footnotes
-
S. Rijkenberg, S. C. van Steen, J. H. DeVries, et al. Accuracy and reliability of a subcutaneous continuous glucose monitoring device in critically ill patients. Journal of Clinical Monitoring and Computing, 2017. https://doi.org/10.1007/s10877-017-0086-z ↩
-
Marek Protus, Barbora Voglová Hagerf, Antonin Jabor, et al. Accuracy and feasibility of real-time continuous glucose monitoring in early postoperative intensive care after liver transplantation. Therapeutic Advances in Endocrinology and Metabolism, 2026. https://doi.org/10.1177/20420188251405372 ↩
-
James Ling, Jack K. C. Ng, Juliana C. N. Chan, et al. Use of Continuous Glucose Monitoring in the Assessment and Management of Patients With Diabetes and Chronic Kidney Disease. Frontiers in Endocrinology, 2022. https://doi.org/10.3389/fendo.2022.869899 ↩
-
Aneta Hásková, Lucie Radovnická, Lenka Petruželková, et al. Real-time CGM Is Superior to Flash Glucose Monitoring for Glucose Control in Type 1 Diabetes: The CORRIDA Randomized Controlled Trial. Diabetes Care, 2020. https://doi.org/10.2337/dc20-0112 ↩
-
Ramzi Ajjan, David Slattery, Eugene Wright. Continuous Glucose Monitoring: A Brief Review for Primary Care Practitioners. Advances in Therapy, 2019. https://doi.org/10.1007/s12325-019-0870-x ↩
-
Alessandro Rossi, Annieke Venema, Petra Haarsma, et al. A Prospective Study on Continuous Glucose Monitoring in Glycogen Storage Disease Type Ia: Toward Glycemic Targets. The Journal of Clinical Endocrinology & Metabolism, 2022. https://doi.org/10.1210/clinem/dgac411 ↩
-
Vera Weijer, Rob van der Werf, Myrthe van der Haijden, et al. Continuous glucose monitoring in para cyclists: An observational study. European Journal of Sport Science, 2024. https://doi.org/10.1002/ejsc.12220 ↩
Pothos builds longitudinal molecular profiles of individuals: whole-genome sequencing, single-cell RNA sequencing, and plasma proteomics, delivered as a detailed report, every data file, and stored cells. Build your profile.