NORM.INV in Google Sheets: Normal Percentiles and Cutoffs
Use NORM.INV in Google Sheets to turn cumulative probabilities into modeled values, calculate central intervals and distinguish upper-tail requests from percentiles.
Verification status: See the evidence and test scope described in this guide. How formula examples are checked
Quick answer: =NORM.INV(0.9,50,5) has expected result approximately 56.4077578277. It is the 90th percentile of a normal model with mean 50 and standard deviation 5: 90% of modeled values are at or below that cutoff.
Syntax and the probability input
=NORM.INV(x, mean, standard_deviation)
Google names the first argument x; for this inverse calculation, it represents the cumulative probability. The NORM.INV entry points to the individual NORMINV reference.
All three arguments are required. No defaults are documented. Supply a probability strictly between 0 and 1 and a positive standard deviation for a finite, nondegenerate normal quantile. Enter 90% or 0.9 for the 90th percentile, not 90.
The result has the same units as the mean and standard deviation. The inverse transforms a probability into a value; the forward normal CDF transforms a value into a probability.
Worked example: modeled processing-time percentiles
Assume a hypothetical normal model with mean 50 minutes and standard deviation 5 minutes. Put those values in D1 and E1. Enter 0.1, 0.5, 0.9 in A2:A4.
In B2, enter and fill down:
=NORM.INV(A2,$D$1,$E$1)
| Probability | Expected cutoff in minutes |
|---|---|
| 0.1 | 43.5922421723 |
| 0.5 | 50.0000000000 |
| 0.9 | 56.4077578277 |
The median equals the mean in this symmetric model. The 10th and 90th percentiles are equally far from 50. Those checks help catch reversed probabilities or a mistaken standard deviation.
These are model-derived percentiles. They are not the same operation as finding a percentile directly from a finite list of recorded times.
Verify the inverse with a forward calculation
For the 90th-percentile cutoff in B4:
=NORM.DIST(B4,$D$1,$E$1,TRUE)
Expected result: approximately 0.9, subject to numerical precision. Google’s normal distribution reference supplies the forward CDF contract.
Keep B4’s full stored result when checking the round trip. Typing a rounded display value such as 56.41 changes the input and will not reproduce exactly the same probability. Round only at the stage where the reporting or operational rule requires it.
Advanced application: a central 90% model interval
A central 90% interval leaves 5% in each tail. Use:
=NORM.INV(0.05,50,5)
=NORM.INV(0.95,50,5)
Expected endpoints are 41.7757318652 and 58.2242681348 minutes. The modeled probability between them is 0.95 − 0.05 = 0.90.
Do not call this automatically a 90% confidence interval for the unknown mean. It describes the central portion of the specified distribution of observations. A confidence interval for a parameter requires a separate sampling model and treatment of estimation uncertainty.
For a cutoff exceeded by only 10% of modeled observations, use the cumulative probability 0.90, not 0.10. The right-tail question has to be converted to its left-tail complement before calling this inverse.
Mathematical interpretation and units
The normal quantile can be written as mean plus standard deviation times a standard-normal quantile. For this example, 50 + 5 × 1.2815515655 gives approximately 56.4077578277.
NIST’s normal-distribution reference describes the normal density, CDF and numerically computed percent-point function. These mathematical definitions support the independent checks; they are not native Sheets execution evidence.
If the model is converted from minutes to seconds, convert both the mean and standard deviation by the same factor. A mean in minutes paired with spread in seconds is not a valid representation of the intended model.
Errors and troubleshooting
Check probability scale first, then positive spread and argument order. Probabilities at 0 or 1 correspond to unbounded normal tails mathematically, so use interior probabilities for finite cutoffs. A negative result can be mathematically valid for a normal model, even if the modeled real-world quantity cannot be negative; that may signal an unsuitable model.
The short NORMINV reference does not enumerate all native errors or coercions. This guide does not borrow the error tokens of a different function or product. It also does not establish extreme-tail precision, blank handling or array broadcasting.
Availability and verification
Google links NORM.INV to NORMINV with matching three-argument syntax. Original introduction date, first version and last functioning numbered release were not established for rolling Sheets. Other applications’ defaults and error behavior should be checked separately.
NOT_TESTED_IN_GOOGLE_SHEETS. The individual Google references, dotted alias entry and NIST definition were inspected on October 17, 2026. Quantiles and forward round trips were independently calculated; no native session or screenshot is claimed.