How this instrument works
The standard normal cumulative distribution function, Φ(z), answers 'what fraction of a standard normal distribution lies at or below this z-value?' The inverse normal distribution — also called the probit function or the quantile function — runs that question backwards: given a cumulative probability p, what z-value produces it? This is the calculation behind every 'find the z-score for the 95th percentile' or 'what z cuts off the top 2.5%' question in introductory statistics, and it's the mechanism underneath building confidence intervals, setting control limits, and looking up critical values.
Neither the normal CDF nor its inverse has an elementary closed-form formula — both are defined by an integral that can't be written using ordinary algebra, sines, or exponentials alone. This calculator uses Peter J. Acklam's 2003 rational approximation of the inverse standard normal CDF, a well-known, widely implemented numerical method accurate to roughly 1.15×10⁻⁹ relative error — far beyond what any practical use case needs, and cross-checked here against the standard, universally published normal-quantile constants (1.6449, 1.9600, 2.3263, 2.5758) already used throughout this site's other statistics instruments.
This instrument takes any cumulative probability p you specify directly and returns a single z-value. It answers a more general question than this site's critical-value calculator, which is framed specifically around a significance level (alpha) and returns both a one-tailed and a two-tailed critical z at once for that alpha — a narrower, hypothesis-testing-oriented framing built on the same underlying approximation. Use this calculator when you have a specific cumulative probability in mind; use critical-value when you're working directly with a significance level.
- Enter the cumulative probability you want to invert into Cumulative probability, p = Φ(z), as a decimal between 0 and 1.
- Read z such that Φ(z) = p beneath — the standard normal z-value that produces your chosen probability.
- p = 0.5 always returns z = 0, the exact center of the standard normal distribution.
- p above 0.5 returns a positive z; p below 0.5 returns a negative z, by the distribution's symmetry.
- Use this for percentile lookups (p = 0.90 for the 90th percentile) or tail cutoffs (p = 0.975 for the value cutting off the top 2.5%).
Worked example — the 97.5th percentile cutoff
Building a 95% confidence interval requires knowing the z-value that cuts off the top 2.5% of a standard normal distribution, leaving 97.5% of the distribution at or below it — the classic 'z = 1.96' figure quoted throughout introductory statistics. Enter 0.975 into Cumulative probability, p = Φ(z).
z such that Φ(z) = p reads 1.959964 (more precisely, 1.9599639845400545), matching the standard published normal-quantile reference to within a hair's width of Acklam's approximation's own tiny error margin. By the standard normal distribution's symmetry about zero, entering p = 0.025 instead returns the exact negative mirror, z ≈ −1.959964 — the lower cutoff for that same 95% central interval.
Questions
What does Φ⁻¹(p) actually mean?
Φ (capital phi) is standard notation for the standard normal cumulative distribution function — Φ(z) gives the probability that a standard normal random variable is less than or equal to z. Φ⁻¹, the inverse, runs that relationship backwards: given a probability p, Φ⁻¹(p) is the z-value at which the cumulative probability reaches exactly p. It's also called the probit function or the quantile function of the standard normal distribution.
Why doesn't this have an exact algebraic formula?
The normal distribution's cumulative probability is defined by an integral of the bell-curve function that has no closed-form expression in terms of ordinary algebra, and neither does its inverse. In practice this calculator (and essentially every statistics tool, textbook table, and software package) uses a numerical approximation instead — here, Peter J. Acklam's widely used 2003 rational-function method, accurate to about 1.15×10⁻⁹ relative error, which is precise enough that the difference from a true closed-form answer is never practically detectable.
How is this different from this site's critical-value calculator?
This calculator takes any cumulative probability p you specify and returns the single matching z-value directly — a general-purpose inverse-CDF lookup. Critical-value is framed specifically around hypothesis testing: you give it a significance level alpha, and it returns both the one-tailed critical z (Φ⁻¹(1−alpha)) and the two-tailed critical z (Φ⁻¹(1−alpha/2)) at once, since a hypothesis test typically needs both framings depending on whether it's one- or two-sided. Both tools share the same underlying Acklam approximation.
What p-value should I enter to find a standard z-score table value?
Enter the exact cumulative probability shown in the z-table you're trying to reproduce. For the common 90%, 95%, and 99% confidence-interval cutoffs, that's p = 0.95, 0.975, and 0.995 respectively (each one is the two-tailed confidence level's upper cutoff, since half the leftover probability sits in each tail); this calculator will return the familiar 1.6449, 1.9600, and 2.5758 figures.
What happens if I enter p = 0 or p = 1?
Neither is accepted — the true inverse normal distribution approaches negative infinity as p approaches 0 and positive infinity as p approaches 1, so there's no finite z-value at the exact endpoints. The calculator requires p to sit strictly between 0 and 1 (in practice, a small margin short of each endpoint) and will prompt you to adjust the value if it falls outside that range.