NORM.INV in ONLYOFFICE: Find a Normal Percentile in the Original Units

Create a sheet named PackingModel. These are assumed model parameters, not measurements or a fitted dataset. Put Parameter in A1 and Value in B1, then enter:

Keep this input block unchanged while adding formulas in D2 onward. Enter 0.95 as a number, or 95% with percentage formatting, rather than the number 95. The numeric value 0.95 and the percentage 95% represent the same probability; the plain number 95 does not.

A normal model is symmetric and continuous. It is an assumption about a distribution, not proof that actual packing data are normal, independent or stable. The parameters must be chosen or estimated separately. In particular, a standard deviation is not a standard error of a sample mean; substituting one for the other changes the modeled quantity.

The official reference requires three arguments. Probability must be strictly between 0 and 1, and standard_dev must be positive. Mean and the resulting threshold use the original measurement units.

This is an inverse CDF. If p=0.95, the returned q satisfies P(X≤q)=0.95 mathematically. It does not return a density, an upper-tail probability or a confidence level for an estimated mean. A percentile of a modeled individual fill is a different quantity from a confidence interval for the process mean.

D3 equals the mean because normal distributions are symmetric. D2 and D4 are equally far above and below 500. D7 is a mathematical round-trip check; real calculation precision and display rounding may prevent exact equality. Compare with a tolerance suited to the intended units, not a requirement that two floating-point cells are identical.

For an upper-tail allowance of 5%, use p=1−0.05=0.95. Supplying 0.05 instead chooses the low threshold on the opposite side. For a central 95% interval, split the remaining 5% between both tails: use p=0.025 and p=0.975, which produce D6 and D5. Using 0.05 and 0.95 instead gives a central 90% interval.

Place 0.95 in F2 as the desired central mass. =NORM.INV((1-F2)/2,B2,B3) and =NORM.INV(1-(1-F2)/2,B2,B3) reproduce the central limits. These bounds describe the assumed distribution of individual fills. They are not guaranteed process limits and do not by themselves establish a statistical confidence interval.

The independent identity is q=mean+standard_dev×Φ⁻¹(p). For p=0.95, the standard quantile is approximately 1.64485362695147; multiplying by 10 and adding 500 explains D2. If the assumed standard deviation doubles to 20 while the mean remains 500, the same percentile is about 532.897072539029. This controlled change doubles the distance from the mean, not the mean itself.

The tagged calculation explicitly rejects p≤0, p≥1 and standard_dev≤0 with #NUM!. Thus =NORM.INV(0,500,10), =NORM.INV(1,500,10) and =NORM.INV(0.95,500,0) are source-predicted #NUM!, not finite thresholds. Mathematical normal quantiles tend to infinities at the endpoints, but this function does not expose those infinities as accepted results.

Do not silently clamp a wrong percentage into the valid interval. The number 95 is not 95%; inspect the cell value and formula before masking an error. Nonnumeric conversions can fail and existing argument errors propagate. These examples deliberately avoid ambiguous date, blank and text coercion.

NORMINV is the legacy three-argument spelling. NORM.INV directly inherits its calculation in the pinned source, including the bounds and gaussinv-based transformation. The code selects the first element of explicit array arguments in this path; these scalar examples do not establish spill behavior. Naming equivalence here does not prove every client or imported workbook is interchangeable.

Extreme probabilities close to 0 or 1 are numerically sensitive. A rounded stored probability of 1 reaches an invalid endpoint even if an earlier unrounded value was valid. Keep probabilities numeric and unrounded during calculations, and inspect actual target-build precision before relying on rare-tail thresholds.

These notes were checked on October 18, 2026 against the linked official help and SDKJS tag v10.0.0.169, commit 3679e0375a758d7dc59d31337a45461be3525ed8. The original introduction date and final supported-version endpoint are not established. Current help is not a complete historical compatibility table; Desktop Editors, Docs and mobile releases need separate verification.

Neither the ONLYOFFICE application nor its calculation engine was executed for this guide. Numerical targets in this guide were independently calculated from normal-distribution mathematics; source observations come from reading the pinned implementation. Rounded targets are not promises of identical displayed digits. There are no new native screenshots or observed workbook results.

Examples use English names and comma separators. Adapt separators to the actual editor without changing argument order. Scalar examples do not establish dynamic-array spill, imported-workbook parity, every coercion rule or behavior in older releases. Retain unrounded probabilities for later calculations and format cells for display rather than replacing stored values with rounded text.

• Official NORM.INV help, checked October 18, 2026.
• Pinned calculation and name definitions.
• Pinned normal-distribution helpers.
• NORM.DIST.
• NORM.S.DIST.
• NORM.S.INV.
• NORMSDIST.