A critical value is the number a hypothesis test compares its statistic against — the line that separates 'not enough evidence' from 'reject the null hypothesis.' Before spreadsheets and statistical software made p-values instant to compute, students and researchers looked these numbers up in printed tables at the back of a textbook; this calculator replaces that table lookup with an exact inverse-CDF calculation for any alpha and any degrees of freedom, across the four distributions most hypothesis tests rely on.
What a critical value actually represents
Every hypothesis test reduces your data to a single test statistic, then asks: how extreme would this statistic need to be before it's implausible under the null hypothesis? The critical value is that threshold, set so that the probability of crossing it purely by chance equals your chosen significance level (α). It's mathematically the same information as a p-value, just computed in the opposite direction — a p-value asks "how extreme is my statistic," while a critical value asks "how extreme would a statistic need to be." Many textbooks and lab courses still teach the critical-value approach first because it makes the decision rule visible before you ever see real data.
Choosing one-tailed vs. two-tailed
Two-tailed is the standard, conservative default: it tests for a difference in either direction without assuming ahead of time which way the effect will go, so the rejection region — and the critical value — sits at α/2 in each tail. A one-tailed test is only valid if you committed to a specific direction (e.g., "the new process reduces defects") before collecting data; switching to one-tailed after seeing which way the data leaned is a common and serious mistake, since it effectively doubles your real Type I error rate. When in doubt, use two-tailed.
Why four different distributions
Z assumes you know the population standard deviation exactly (or your sample is large enough that the distinction barely matters). t relaxes that — it's what you use when the standard deviation is estimated from the sample itself, and it has heavier tails to compensate for that extra uncertainty, converging to the Z distribution as degrees of freedom grow large. Chi-square applies to counts and variances rather than means, and because it's bounded below by zero and right-skewed, its critical value is conventionally reported for the upper tail only. F is the ratio of two variances (or two mean-squares in ANOVA) and needs two degrees-of-freedom values — it isn't symmetric, so a two-tailed F test genuinely needs two distinct critical values rather than a single ± pair.