NORMSINV in Apple Numbers: Z-Score Quantiles

NORMSINV finds the standardized threshold for a requested cumulative percentage. Its most common interpretation is Φ⁻¹(p): the z where the standard normal CDF reaches probability p. It is different from NORMSDIST, which goes from z to p, and from NORMINV, which returns a measurement in original units after applying the supplied mean and population SD. [A1, A10]

=NORMSINV(probability)

Only one argument is documented. There is no mean argument, no stdev argument and no optional mode. The standard-normal distribution underlying the function has mean 0 and standard deviation 1. Percent labels such as 95% should correspond to underlying value 0.95; entering number 95 is outside the documented probability domain. Negative output z-scores are legitimate for probabilities below one half. [A1, A2]

The strict exclusion of p=0 and p=1 is mathematical as well: the standard normal has infinite tails. Apple’s help does not promise automatic clipping, an error message spelling or a fixed maximum finite precision for probabilities very near an endpoint. [A1, C1]

Create a Numbers table named ZCutoffs, enter Cumulative probability in B1 and z cutoff in C1. Type the numeric probabilities in rows 2–6, enter =NORMSINV(B2) into C2, and copy down. Format C to show several decimals, especially for probabilities of 0.025 and 0.975. [A1, A3]

The central 95% band of the standard-normal distribution is bounded by approximately -1.959964 and +1.959964. The middle of the distribution corresponds to p=0.5 and z=0; a probability of 0.05 maps to a negative z around -1.644854. These are purely distributional quantiles, not automatically confidence intervals around a sample mean. [C1]

If you put probabilities in B and the formula in a separate table, use a qualified reference such as ZCutoffs::B2 instead of guessing an Excel worksheet reference. [A3]

Apple gives =NORMSINV(0.89) → 1.22652812003661. Independent inverse-standard-normal mathematics returns the same value to the documented digits. Checking the forward direction with =NORMSDIST(NORMSINV(0.89)) yields approximately 0.89, subject to floating-point rounding. This is a mathematical check—not proof of a formula evaluation on Mac, iPad, or iPhone. [A1, A10, C1]

Standard normal upper-tail cutoff

If the upper tail probability is 5%, the cumulative probability to its left is 1−0.05=0.95. Therefore:

=NORMSINV(1-0.05)

Expected 1.64485362695. Using NORMSINV(0.05) instead produces -1.64485362695, the opposite end of the normal curve. The distinction matters for one-sided thresholds. [C1]

Two-sided 95% central band

=NORMSINV(0.025)
=NORMSINV(0.975)

Expected approximately -1.95996398454 and +1.95996398454. The area between the two is 95%. If you want an actual confidence interval around an estimated population mean, additional sample-size and sampling-uncertainty calculations are necessary; these two quantiles alone do not provide that interval. [C1]

Convert a standard z-score into a measurement cutoff

For a normal measurement model with mean 50 and population SD 10:

=50+10*NORMSINV(0.95)
=NORMINV(0.95,50,10)

Both independently produce 66.4485362695 in measurement units. The first performs explicit arithmetic after obtaining z; the second uses Apple’s documented general inverse function. Apple notes that NORMINV(p,0,1) agrees with NORMSINV(p), so this scaled relationship is mathematically consistent with its definition. [A10, C1]

Inspect round-trip consistency after entering parameters

=NORMSDIST(NORMSINV(0.025))
=NORMSDIST(NORMSINV(0.975))

Expected about 0.025 and 0.975 respectively. Round-trip checks reveal swapped probabilities and wrong argument interpretation, although they cannot prove an assumed normal model is appropriate for real observations. [C1]

Apple lists input domains for these functions but does not enumerate the precise displayed error token for every invalid input, version, imported document, or type conversion. Therefore the examples below describe the cause and corrective action, not an invented #... error string. [A1, A4]

When communicating thresholds, state whether p is left-cumulative or right-tail, and whether the result is in z-score units or original units. Otherwise a numerically correct result can be applied to the wrong side of a distribution. [C1]

The worksheets above assume formulas and their source data live in the same Numbers table, so B2 identifies the first input cell under a row-1 header, and $B$2 stays fixed if copied. Unlike a one-sheet-only model, Numbers documents multiple named tables per sheet. If you move the calculation to another table, the cross-table pattern is ZCutoffs::B2 (when the source table is named ZCutoffs). A sheet-qualified form is Sheet 2::ZCutoffs::B2. These conventions come from Apple’s cell-reference guide, not an Excel workbook model. Use the Formula Editor to select the actual cell and avoid accidental name collisions. [A3]

The worksheet in this tutorial uses scalar formulas per row, not a dynamic-array spill convention. Enter row-1 headers first, enter numeric values in source columns, put the formula into the indicated output cell, then fill it down. If copy/fill moves a parameter that should be fixed, use a fixed reference like $C$2, or explicitly refer to the same source row. Numbers documents the difference between A1 and $A$1; copying a formula may change the former but not the latter. [A3]

This page uses the English function name NORMSINV and English-language examples with a period as decimal point and a comma between arguments. Apple’s Brazilian Portuguese 15.4 help for the same function instead titles the corresponding function INV.NORMP and demonstrates syntax with the locale-appropriate separator; an independently checked example is =INV.NORMP(0,89). This is a verified localized name tied to the same official help identifier. [A9]

Copying an English formula verbatim to a Portuguese-Brazil installation can therefore fail or be interpreted differently. Inspect the insertion suggestion in that installation’s Functions Browser and match the spreadsheet’s language/region settings. In particular, a comma in 0,89 is a decimal separator in the Brazilian example while semicolons separate arguments. The B2:C6 colon in a range is neither of those separators. Apple documents the general Functions Browser on Mac/iPhone/iPad. [A5, A9]

The reference version used here is Apple Formulas and Functions Help 15.4 (English, checked 2026-10-13 UTC). Apple lists the 15.4 Numbers release on Mac and iPhone, and the current release notes specify macOS 26 or later for Numbers 15.4 on Mac and iOS 26 or later on iPhone. Those are app-level requirements, not proof that this particular function was tested on either device, available in every earlier version, or available identically in Pages, Keynote and iCloud. [A5, A6, A7]

These functions are documented in Apple Formulas and Functions Help 15.4; the release notes do not establish the earliest version that introduced this specific function. Do not convert the current documentation-version number into an unsupported first-supported-version claim. [A1, A6, A7]

A different spreadsheet may use a similarly named normal-distribution function or different punctuation and defaults. This article makes no claims about Excel, Google Sheets, or other spreadsheet engines, and it does not silently treat their behavior as evidence for Numbers. Where the function page omits a property (for example exact error strings, limits for extreme finite values, coercion of blank/text cells, or internal floating-point precision), this guide deliberately marks it unverified rather than inventing it. [A1, A4]

Validation boundary: Every displayed numeric expectation was independently calculated in a Python/SciPy statistical calculation environment using the displayed formulas and inputs; this is not native execution in Apple Numbers. The official examples were also compared with the published values. No screenshot, Numbers file, import/export conversion, formula-error display, or cross-platform execution test was produced. Roundoff and the number of decimals shown in a Numbers result cell depend on numeric calculation/display choices; compare sufficiently many digits before assuming a mistake. [C1]

• NORMSINV: exact syntax and published examples — this page’s authoritative English reference. [A1]
• NORMDIST — general normal cumulative probability or density selected explicitly by mode. [A10]
• NORMINV — inverse of the general normal CDF. [A10]
• NORMSDIST — standard normal cumulative probability (not a density). [A10]
• NORMSINV — inverse of the standard normal CDF. [A10]
• PHI — standard normal density at one z-score. [A10]

• [A1] Apple — NORMSINV (Formulas and Functions Help 15.4) — function name, purpose, signature, type/domain, named examples, help-linked relationships; checked 2026-10-13.
• [A2] Apple — Types of arguments and values — Apple definitions of numeric/modal arguments (where the individual function specifies a modal choice); checked 2026-10-13.
• [A3] Apple — Refer to cells in formulas — B2, B2:B6, table and sheet :: notation, relative and absolute references; checked 2026-10-13.
• [A4] Apple — Review formula errors and best practices — general formula debugging, not a source of invented exact errors; checked 2026-10-13.
• [A5] Apple — Formulas and Functions Help welcome — Functions Browser and iWork app context; checked 2026-10-13.
• [A6] Apple — What’s new in Numbers on Mac — Numbers 15.4 release and OS requirement; checked 2026-10-13.
• [A7] Apple — What’s new in Numbers on iPhone — Numbers 15.4 release and OS requirement; checked 2026-10-13.
• [A8] Apple — New functions in Numbers on iPad — names PHI specifically among current new statistical functions; checked 2026-10-13.
• [A9] Apple — Portuguese-Brazil help for INV.NORMP — function translation and locale-specific formula example; checked 2026-10-13.
• [A10] Apple — linked normal-distribution family pages — independently checked each of the five named official Apple help URLs on 2026-10-13; these establish relative purpose without asserting spreadsheet cross-engine compatibility.
• [C1] Mathematical calculations using independent Python/SciPy (scipy.stats.norm) and standard normal PDF/CDF identities; using the displayed example values. Never a native Numbers run.