How this instrument works
A critical value marks the boundary of a hypothesis test's rejection region: the point on the standard normal distribution beyond which a computed test statistic is considered too extreme to have plausibly arisen by chance alone, at a chosen significance level, alpha. This calculator inverts the standard normal cumulative distribution function to find that boundary for any alpha you enter, rather than being limited to the handful of round numbers (0.10, 0.05, 0.01) most textbook tables print.
It reports two related values. The one-tailed critical z answers 'how far out, on one side only, does the extreme alpha fraction of the distribution begin' — used when a test only cares about deviation in one direction. The two-tailed critical z instead splits alpha in half, putting half in each tail, since a two-tailed test flags extreme values on either side as evidence against the null hypothesis — this is the more commonly cited value, and it's why alpha=0.05 famously produces the widely known z=1.96 figure (Phi⁻¹(0.975), not Phi⁻¹(0.95)).
This is a general-purpose lookup utility, not a full hypothesis test: it takes only alpha and returns z, with no sample data, no test statistic, and no accept/reject decision attached. It's also specifically a z-value from the standard normal distribution, not a t-value — it doesn't account for sample size or use the t-distribution's fatter tails, which matter when a population's true standard deviation is unknown and a small sample's own standard deviation is used as a stand-in instead. For that situation, a t-distribution critical value table, not this normal-based calculator, is the appropriate tool.
- Enter your chosen one-tailed significance level into Significance level, alpha (one-tailed tail probability) — a value strictly between 0 and 0.5.
- Read One-tailed critical z for the boundary that puts the entire alpha probability in a single tail.
- Read Two-tailed critical z for the boundary that splits alpha in half between both tails — the more commonly used figure for two-sided tests.
- Common values to check against: alpha=0.05 gives a two-tailed z of 1.96, and alpha=0.01 gives a two-tailed z of about 2.576, both standard textbook constants.
Worked example — the standard 5% significance level
Entering alpha = 0.05 into Significance level, alpha, the instrument computes One-tailed critical z as Phi⁻¹(0.95), which reads 1.644854 — matching the widely published normal-quantile reference value of 1.6448536269514722 that appears throughout introductory statistics texts. This is the boundary used when a test only flags results that are extreme in a single, specified direction.
Two-tailed critical z reads 1.959964, computed as Phi⁻¹(0.975) — the famous 1.96 figure cited constantly across statistics, matching the standard published value of 1.9599639845400545. This is the boundary used for the far more common two-sided test, which flags a result as significant whether it lands unusually far above or unusually far below what's expected, splitting the 5% alpha into 2.5% in each tail.
Questions
Why are the one-tailed and two-tailed values different for the same alpha?
Because they distribute the same total alpha probability differently across the distribution's tails. A one-tailed test puts the entire alpha in a single tail, so the boundary sits closer to the center of the distribution; a two-tailed test splits alpha in half between both tails, which means each individual tail only needs to capture alpha/2 of the probability, pushing that boundary further out. That's exactly why alpha=0.05 gives a smaller one-tailed z (1.6449) than two-tailed z (1.96) — the two-tailed test needs a stricter, more extreme threshold on each side to keep the same overall 5% total.
Should I use a one-tailed or two-tailed critical value?
Use one-tailed when your hypothesis specifically predicts a difference in only one direction and you have a solid reason to ignore the opposite direction entirely (a new drug can only help, never hurt, for instance); use two-tailed whenever a difference in either direction would be meaningful, which is the more common and generally more conservative choice by default. Switching to one-tailed after seeing which direction the data leans, rather than deciding beforehand, is a well-known way to inflate false-positive rates and should be avoided.
Why does this use z instead of a t-value?
This calculator inverts the standard normal distribution, which is the correct choice when the population standard deviation is known, or the sample is large enough that the normal approximation holds well. When the population standard deviation is unknown and must be estimated from a small sample, the t-distribution's fatter tails (which depend on the sample's degrees of freedom) are the statistically correct choice instead, and its critical values run larger than this calculator's z-values for the same alpha — this instrument doesn't compute those, since they require a degrees-of-freedom input this page doesn't take.
How is this different from the site's hypothesis-testing calculator?
This is a general-purpose lookup: enter any alpha in (0, 0.5) and get the matching z-values, nothing more. This site's separate hypothesis-testing calculator is a full decision workflow built around a fixed table of common alpha choices (10%, 5%, 1%) — it takes an actual sample mean, hypothesized mean, standard deviation, and sample size, computes a test statistic, and returns an explicit reject/fail-to-reject decision. Use this page when you just need the critical z-value itself for any alpha; use hypothesis-testing when you have real sample data and want the full test carried out.
How accurate is the critical value this calculator returns?
It's computed via Peter J. Acklam's 2003 rational approximation of the inverse standard normal CDF, which has an empirically measured relative error around 1.15×10⁻⁹ — accurate enough to match published normal-quantile reference tables to at least six decimal places for any alpha in the valid (0, 0.5) range. That's precise enough for essentially any applied statistics use, well beyond the precision most published z-tables print.