T-Distribution Table (PDF)

What Is a T-Distribution Table? Definition and Purpose

The t-distribution table — also called a t-table, t-score table, t-value table, or Student’s t-table — is a reference chart listing critical values of the Student’s t-distribution. Every cell in the table tells you the t-value at which a given percentage of the distribution falls in the tail or tails. You compare your calculated t-statistic against these threshold values to make a decision in hypothesis testing. Scribbr’s t-table guide defines it well: critical values of t define the threshold for significance for certain statistical tests and the upper and lower bounds of confidence intervals.

In plain terms: you run a t-test, get a number, and then check the t-distribution table to find out if that number is big enough to be statistically meaningful. The table does the hard work of the probability calculation so you don’t have to integrate the t-distribution by hand. It is the working student’s best friend in any statistics course — whether you are enrolled at Harvard University, University College London, or a community college running AP Statistics.

The t-distribution table is most commonly used for these tasks, as described by Scribbr: testing whether two means are significantly different (two-sample t-tests), testing whether two variables are significantly related (linear regression or correlation), and calculating confidence intervals for means or regression coefficients. These three use cases cover the vast majority of inferential statistics problems that college students encounter. Hypothesis testing is the context where the t-table gets used most heavily.

Why Does the T-Distribution Exist at All?

The normal distribution (z-distribution) works perfectly when you know the population standard deviation. But in real research, you almost never know the true population standard deviation. You have a sample, you estimate the standard deviation from that sample, and that estimation introduces extra uncertainty. The t-distribution accounts for that uncertainty by having fatter tails than the normal distribution — especially when sample sizes are small.

As Statistical Aid explains, the T-distribution has “fatter tails” than the Z-distribution. This means it assigns higher probabilities to extreme values, reflecting the increased likelihood of observing unusual data points due to the smaller sample size and the estimated standard deviation. The fatter tails are the mathematical acknowledgment that small samples are less reliable than large ones — so you need a larger t-value to claim significance with a small sample than you do with a large one.

As sample size grows, the t-distribution’s tails slim down. By about 30 degrees of freedom, the t-distribution and z-distribution are nearly indistinguishable. This is why textbooks often tell you: use the t-table for small samples, use the z-table when n ≥ 30 and population standard deviation is known. The normal distribution is effectively the limiting case of the t-distribution.

What Makes the T-Distribution Table Unique as a Tool?

Unlike the z-table, which has just one fixed shape, the t-distribution table encodes a whole family of distributions — one for each degree of freedom. Every row in the t-table represents a slightly different bell curve. The row for df = 1 describes a very flat, wide distribution with thick tails. The row for df = 120 describes something nearly indistinguishable from the standard normal. This variability is why the t-table has so many rows: each degree of freedom produces different critical values, and the table lists them all. For a more detailed treatment of distribution families, see the site’s guide on probability distributions.

William Sealy Gosset and the Origin of Student’s t-Distribution

The t-distribution has one of the most interesting origin stories in the history of statistics. William Sealy Gosset (1876–1937) was a chemist and statistician employed by the Guinness Brewery in Dublin, Ireland. His job was to use small samples of barley and hops to make inferences about large batches — a problem that pushed him to develop new statistical tools when the existing normal-distribution methods proved inadequate for small samples.

There was a catch. Guinness had a policy prohibiting employees from publishing research, to prevent competitors from gaining insights into their quality-control methods. Gosset found a workaround: he published his 1908 paper introducing the t-distribution in the journal Biometrika under the pseudonym “Student.” That is why the distribution is formally called Student’s t-distribution to this day. As the StatsMasters reference table notes, Gosset’s employer did not allow employees to publish under their own names, which is why the pseudonym stuck.

What Gosset derived was a probability distribution specifically suited to estimating population means when sample sizes are small and the population standard deviation is unknown. The mathematical structure he identified — the ratio of a standard normal variable to the square root of a chi-squared variable divided by its degrees of freedom — is exactly what students encounter when they study the t-statistic formula today. Foundational statistics texts describe it formally: the t-distribution with n degrees of freedom is defined as the distribution of the quotient of a standard normal variable Z and the square root of a chi-squared variable X divided by n.

Ronald Fisher and the Expansion of the T-Table

Ronald Fisher, the British statistician widely regarded as the father of modern statistics, played a central role in popularizing Gosset’s work. Fisher reformulated the distribution in terms of degrees of freedom — a concept that made the table extensible and practical for general research use. Fisher’s 1925 work Statistical Methods for Research Workers, published by Oliver & Boyd in Edinburgh, included early versions of what became the standard t-distribution table format still used today.

Fisher’s concept of degrees of freedom is what makes the t-table’s row structure meaningful. Without the degrees-of-freedom framework, there would be no systematic way to organize the different t-distributions by sample size. Every statistics textbook used at MIT, Oxford University, Stanford University, and London School of Economics today traces its t-table format back to Fisher’s organizational insight. Understanding this history helps students see the t-distribution table not as an arbitrary lookup chart but as the crystallized output of a century of statistical development.

Karl Pearson and Biometrika: The Publication Venue

Karl Pearson, the founding editor of the journal Biometrika, made the publication of Gosset’s 1908 paper possible. Pearson was himself a major figure in early statistics — he developed the chi-square test, introduced the concept of correlation, and founded the first university statistics department in the world at University College London. Without Pearson’s openness to publishing Gosset’s unconventional work in Biometrika, the t-distribution might have languished unpublished. The relationship between Gosset, Fisher, and Pearson shaped the entire landscape of modern inferential statistics, including the t-table that students use today. For more on the chi-square distribution that appears in the t-distribution’s mathematical foundation, see the guide on chi-square tests.

Anatomy of the T-Distribution Table: Columns, Rows, and Values

Before you can use the t-distribution table effectively, you need to understand its physical structure. The layout is not arbitrary — every element of the table has a specific statistical meaning, and misreading any one part leads to the wrong critical value.

The Rows: Degrees of Freedom (df)

The left column of the t-distribution table lists degrees of freedom (df). This is the variable that determines which t-distribution you are working with. Low df values (1, 2, 3) appear at the top; higher values extend down through 120, 200, 500, 1000, and finally ∞ (infinity). Each row represents a different, specific t-distribution. Statistics By Jim is explicit: choose the row of the t-table that corresponds to the degrees of freedom in your t-test.

The final row — typically labeled ∞ or z — lists the critical values of the standard normal distribution. Comparing these to the rows above them shows you exactly how much fatter the t-distribution’s tails are for small samples. The critical value at df = 1, α = 0.05, two-tailed is 12.706. The same alpha at df = ∞ gives 1.960 — the familiar z-score for 95% confidence intervals. That gap between 12.706 and 1.960 tells the whole story of why small-sample statistics needs the t-distribution.

The Columns: Significance Levels (Alpha)

The column headers list significance levels (alpha values) — the probability thresholds that define how much risk of a false positive you are willing to accept. Common alpha values in the t-table are 0.10, 0.05, 0.025, 0.01, 0.005, 0.001, and 0.0005. Some tables separate one-tailed and two-tailed alpha values into different rows of headers; others show a single set and require you to adjust for the tail type. Statistics By Jim advises: choose the column that contains the significance level for your test, and be sure to choose the alpha for a one-tailed or two-tailed test based on your methodology.

When reading a table that shows two-tailed alpha at the top, the values already account for both tails. If you need a one-tailed test at α = 0.05, you use the column labeled 0.10 for two-tailed (because 0.05 × 2 = 0.10 total area in both tails). Or if the table shows one-tailed alpha, use the 0.05 column directly for a one-tailed test and the 0.025 column for a two-tailed test at α = 0.05. This distinction trips up students regularly — always check whether the table header refers to one-tailed or two-tailed values before reading a critical value. Sampling distribution theory explains why this distinction matters mathematically.

The Cell Values: Critical t-Values

The numbers inside the table are the critical t-values. The critical value at a given (df, alpha) combination is the t-score that divides the distribution into the rejection region and the non-rejection region. If your calculated t-statistic exceeds this critical value in absolute terms, you reject the null hypothesis. If it does not, you fail to reject it. Numiqo’s t-distribution tutorial puts it clearly: the critical t-value is the threshold against which you compare your calculated t-statistic.

Two Header Rows: One-Tailed and Two-Tailed

Many t-distribution tables show two rows of headers: one for one-tailed alpha and one for two-tailed alpha. The two-tailed alpha is always exactly twice the one-tailed alpha. So the column for one-tailed α = 0.05 is the same column as two-tailed α = 0.10. T Distribution Table explains: in a one-tail t-table, alpha values start from 0.50, 0.25, 0.20, 0.15 etc.; in a two-tail t-table, they start from 1.00, 0.50, 0.40, 0.30 etc. The structure varies by publisher, so always read the header before using any specific t-table.

Degrees of Freedom: The Single Most Important Variable

Degrees of freedom (df) is the number that determines which row of the t-distribution table you use — and therefore which critical value applies to your test. It is arguably the most important concept for correctly using the t-table. Students who miscalculate degrees of freedom look up the wrong critical value and draw incorrect statistical conclusions. The Statistical Aid guide is direct: degrees of freedom represent the number of independent pieces of information used to estimate a parameter and are arguably the most important factor influencing the shape of the T-distribution.

How to Calculate Degrees of Freedom for Each Test Type

Different t-test designs use different formulas for degrees of freedom. Using the wrong formula is one of the most common errors in undergraduate statistics assignments.

Why Degrees of Freedom Shape the Distribution

The mathematical reason for fatter tails at low df is that with fewer observations, your sample standard deviation is a less reliable estimate of the population standard deviation. The t-distribution accounts for this extra estimation uncertainty by putting more probability mass in the tails — meaning you need a more extreme t-value to achieve statistical significance with a small sample than with a large one.

Think of it this way: if you measure the heights of 3 people and calculate a sample mean, that mean could easily be quite far from the true population mean just by chance. If you measure 300 people, your sample mean will almost certainly be close to the true mean. The t-distribution encodes this logic: small samples need a higher t-threshold for significance, because extreme t-values are more likely to occur by chance when samples are small. This is precisely why the t-table rows for small df show much larger critical values than rows for large df. For a deeper exploration of how sampling variability works, the site’s guide on sampling distributions provides the foundational theory.

What Happens When Your df Is Not in the Table?

Standard t-tables often list df values up to 30 in increments of 1, then jump to 40, 50, 60, 80, 100, 120, and ∞. If your df falls between listed values, StatsMasters advises: if your exact df isn’t listed, use the next smaller df value to be conservative. This is the standard approach for hand calculations. Statistical software handles this automatically by computing the exact critical value for any df, but in an exam setting without software access, rounding down to the nearest listed df is the correct conservative approach.

The Full T-Distribution Table: Critical Values (df 1–∞)

The table below provides critical t-values for both one-tailed and two-tailed tests across the most common significance levels. This is the standard reference table format used in university statistics courses, AP Statistics, A-Level Mathematics and Statistics, and research across medicine, psychology, and social science. The critical values below match those published by institutions including the National Institute of Standards and Technology (NIST), University of Washington Department of Biostatistics, and standard statistical textbooks.

To use this table: find your degrees of freedom (df) in the left column, then move across to the significance level column for your test. The number at the intersection is your critical t-value. NIST’s handbook confirms: the t-table can be used for both one-sided and two-sided tests using the appropriate value of alpha.

These critical values match the published tables at University of Baltimore Statistical Tables and the NIST Engineering Statistics Handbook. For the extended table with additional significance levels (including 0.001 and 0.0005) and additional df values, the University of Washington Biostatistics t-tables provide a comprehensive printable reference used widely in medical and social science research.

One-Tailed vs Two-Tailed Tests: Which Column to Use

Choosing between a one-tailed and a two-tailed test is one of the most consequential decisions in hypothesis testing — and it determines which column of the t-distribution table you use. Getting this wrong leads to a critical value that does not match your hypothesis structure, potentially causing you to reject or fail to reject the null hypothesis incorrectly. As Numiqo explains, a two-tailed test is used when deviations in either direction are important, while a one-tailed test is used when you are only interested in deviations in one specific direction.

Why Two-Tailed Tests Have Higher Critical Values

A two-tailed test at α = 0.05 puts 2.5% of probability in each tail, requiring a more extreme t-statistic to reject the null than a one-tailed test at the same alpha (which puts all 5% in one tail). So the two-tailed critical value is always larger than the corresponding one-tailed value for the same alpha and df. At df = 20, the two-tailed α = 0.05 critical value is 2.086. The one-tailed α = 0.05 critical value is only 1.725. This means it is harder to achieve significance with a two-tailed test — which makes intuitive sense because you are being agnostic about direction.

If you had a theory that a new drug would reduce blood pressure, you would use a one-tailed test (lower tail). If you just wanted to know whether the drug changed blood pressure at all, you would use a two-tailed test. The one-tailed test is more powerful when you correctly predicted the direction — but it is also riskier, because if the effect goes the other way, you will miss it entirely. This is why T Distribution Table advises: if you are in doubt about whether to use a one-tailed or two-tailed test, it is better to go with the two-tailed test generally. For guidance on formulating directional hypotheses correctly before choosing a tail type, the site’s hypothesis testing guide walks through this decision in depth.

The Relationship Between Alpha and Tail Type in the Table

StatsMasters gives a precise rule: for a one-tailed test at α = 0.05, use the column labeled α = 0.05 directly. For a two-tailed test at α = 0.05, use the column labeled α = 0.025, since 0.05 ÷ 2 = 0.025. This applies to tables that label columns by tail area rather than total alpha. Different publishers format their tables differently, so always read the column headers carefully rather than assuming the format matches what you have seen before. This minor formatting difference between tables is the source of many avoidable errors in undergraduate statistics work.

How to Read the T-Distribution Table: Step-by-Step

Reading the t-distribution table correctly requires following a consistent process. Every step matters. Rushing through any one step risks pulling the wrong critical value and making an incorrect inference. Here is the complete procedure, consistent with how Statistics By Jim, Scribbr, and MedCalc describe it.

Using the T-Table for Confidence Intervals

The t-distribution table is not just for hypothesis testing. It is equally essential for constructing confidence intervals when the population standard deviation is unknown. A confidence interval gives a range of plausible values for the population mean based on your sample data. The t-critical value from the table defines how wide that interval is. The MedCalc t-table guide explains: critical t-values are used to establish the upper and lower bounds of confidence intervals.

The Confidence Interval Formula

The confidence interval formula multiplies the critical t-value by the standard error of the mean. This product is the margin of error. Add it to the sample mean for the upper bound; subtract it for the lower bound. The result is a range within which you are (1 – alpha) × 100% confident the true population mean falls.

Worked Confidence Interval Example

Why a Wider Interval Means More Confidence

A 99% confidence interval is wider than a 95% confidence interval because you use a larger critical t-value (from the α = 0.01 column instead of α = 0.05). The trade-off is clear: more confidence means less precision. You can be nearly certain your interval contains the true mean, but only by making the interval so wide that it becomes less informative. This is the fundamental tension in confidence interval construction, and it is entirely governed by the t-table lookup. Confidence intervals are covered in depth on the site for students who need a full treatment of this topic.

Choosing the confidence level is a research design decision, not a statistical one. 95% (α = 0.05) is the default in most academic fields because it balances confidence and precision adequately for most research purposes. Medical and pharmaceutical research often uses 99% (α = 0.01) when the stakes of a false conclusion are high. Exploratory work might use 90% (α = 0.10) to generate hypotheses without requiring high certainty.

T-Table vs Z-Table: When to Use Which

One of the most persistent questions in introductory statistics is when to use the t-distribution table versus the z-table (standard normal distribution table). The answer is simpler than most students expect once you understand the underlying logic. T Distribution Table states the rule directly: a Z Table is used when the population standard deviation and mean are known, whereas a T Table is used when the T score is calculated without knowledge of the mean and the population standard deviation.

The Large-Sample Convergence Rule

For large samples — generally n ≥ 30 — the t-distribution and z-distribution produce nearly identical critical values. At df = 120, the two-tailed α = 0.05 critical value is 1.980, compared to the z-value of 1.960. The difference is so small that using either table produces the same practical conclusion. This is why textbooks often say “use the t-table when n is small” — it is shorthand for “when the difference between t and z critical values is large enough to matter.” For practical purposes in most university statistics courses, using the t-table is always safe and appropriate, even for large samples — the t-table gives a slightly more conservative critical value that reduces false positive risk. Student’s t-distribution is explored more formally for those who want the mathematical proof of this convergence.

The One Scenario Where You Must Use the T-Table

The situation where using the z-table instead of the t-table causes a real error is with small samples and unknown σ. A researcher with n = 10 who mistakenly uses z = 1.960 as their two-tailed critical value at α = 0.05 is using a threshold that is too lenient. The correct critical value from the t-table at df = 9, two-tailed α = 0.05 is 2.262. Using 1.960 instead of 2.262 means rejecting the null hypothesis at effect sizes that should not reach significance — a false positive inflated by using the wrong table. This is not a theoretical concern — it has contributed to replication failures in psychological and medical research that relied on underpowered designs with inappropriate statistical tools.

Worked Examples Across Common Statistical Tests

The most effective way to master the t-distribution table is to work through problems from different test designs. The examples below cover the three most common t-test scenarios that appear in university statistics courses, AP Statistics, and A-Level Mathematics. Each problem follows the full seven-step procedure from Section 7 so you can see exactly how every decision maps to a table lookup. For additional quantitative practice problems similar to these, the simple linear regression guide covers the t-test for regression coefficients in depth.

Example 1: One-Sample T-Test (AP Statistics Style)

Example 2: Independent Two-Sample T-Test

Example 3: Paired T-Test (Before/After Design)

Common Mistakes When Using the T-Distribution Table

Errors with the t-distribution table are remarkably consistent across students and universities. The following mistakes appear in undergraduate statistics labs, AP exam free-response sections, and A-Level statistics papers with enough regularity that exam boards explicitly flag them in marking guidance. Knowing them in advance is the fastest way to protect your marks.

Mistake 1: Using n Instead of n – 1 for df

This is the single most common error. A student with n = 20 looks up df = 20 in the t-table instead of df = 19. The critical value for df = 20, two-tailed α = 0.05 is 2.086. For df = 19, it is 2.093. In this case, the practical impact is small, but the conceptual error is significant and always loses marks. Degrees of freedom for a one-sample t-test are always n – 1, not n. Write this formula down before approaching the table every time. As Statistical Aid notes, df = n – 1 where n is the sample size — this is the standard formula for a single-sample or paired design.

Mistake 2: Using the Wrong Tail Column

A student running a two-tailed test looks up the one-tailed α = 0.05 column and gets a critical value of 1.725 (for df = 20) instead of the correct two-tailed critical value of 2.086. Using 1.725 makes the test too easy to pass — the student rejects the null more readily than justified. The error usually stems from not checking the table header and assuming all columns represent the same tail type. Statistical Aid emphasizes: make absolutely sure you are using the correct column for one-tailed or two-tailed tests. A common mistake is using the wrong column, leading to incorrect conclusions.

Mistake 3: Forgetting That Two-Tailed Critical Values Are Symmetric

For a two-tailed test, the rejection region is symmetric: reject the null if t < –t* OR t > +t*. Students sometimes only check if their t-statistic is greater than the positive critical value and miss cases where a large negative t-statistic also falls in the rejection region. Both tails count. If your t-statistic is –2.5 and your critical value is ±2.086, you reject the null — even though –2.5 is less than 2.086 in absolute terms? No — in absolute value, 2.5 > 2.086. Always use the absolute value of your t-statistic when comparing to a two-tailed critical value.

Mistake 4: Not Rounding Down for Unlisted df Values

A student has df = 37. The t-table lists df = 30 and df = 40 but not df = 37. The correct approach is to use df = 30 (round down), which gives a slightly larger critical value and produces a more conservative test. Students who round up to df = 40 get a smaller critical value, making it slightly easier to reject the null — this is anti-conservative and statistically improper. Always round down when your exact df is not listed. StatsMasters confirms this explicitly.

Mistake 5: Confusing the T-Statistic with the Critical Value

Students sometimes compare the p-value to the t-statistic, or compare the t-statistic to the alpha level, rather than comparing the t-statistic to the critical value and the p-value to alpha. The correct pairs are: t-statistic compared to critical t-value, and p-value compared to alpha. These are two different ways of making the same decision, but mixing the pairs produces nonsense. If you use software that gives you a p-value, compare it directly to alpha — you do not need the t-table at all. The t-table is specifically for when you have the t-statistic and need to determine the rejection region manually.

T-Distribution in SPSS, Excel, R, and Python

While the printed t-distribution table remains the foundational reference for understanding the statistical logic, modern research is conducted using software. Knowing how to obtain t-critical values and p-values from statistical software is an essential practical skill for any student or researcher who works with data. Inferential statistics in academic and professional settings today is almost always conducted through these tools rather than by hand lookup.

Microsoft Excel

Excel provides two key functions for the t-distribution. T.DIST(x, df, cumulative) returns the cumulative probability up to a given t-value. T.INV(probability, df) returns the t-value corresponding to a given left-tail probability — this is the inverse function you use to find critical values. For a two-tailed critical value: =T.INV.2T(0.05, 20) returns 2.086 for df = 20 at α = 0.05 two-tailed. For one-tailed: =T.INV(0.95, 20) returns 1.725. Excel also has T.TEST(array1, array2, tails, type) for running full t-tests directly from your data. The site’s guide on statistical calculations in Excel provides additional function-level guidance for students learning quantitative analysis in spreadsheets.

IBM SPSS Statistics

SPSS runs all three major t-test types through the Analyze → Compare Means menu. The output table provides the t-statistic, degrees of freedom, and two-tailed p-value automatically. You do not need to look up a critical value manually — you simply compare the “Sig. (2-tailed)” value to your chosen alpha. If Sig. < α, the result is statistically significant. SPSS is the software of choice in psychology, sociology, healthcare research, and business analytics programs at universities across the United States and United Kingdom. It handles Levene’s test for equal variances automatically, which informs whether to use the standard two-sample t-test or Welch’s t-test in your SPSS output interpretation.

R Statistical Computing Environment

R provides the t.test() function for all t-test designs. The output includes the t-statistic, degrees of freedom, and p-value. For critical values: qt(0.975, df=20) returns the two-tailed critical value at α = 0.05 for df = 20 (the 0.975 quantile, since 0.05 is split across both tails). pt(t_value, df=20) returns the cumulative probability, from which you derive the p-value. R is the statistical language of choice for quantitative research in economics, epidemiology, and data science at institutions like Johns Hopkins Bloomberg School of Public Health, Harvard T.H. Chan School of Public Health, and Imperial College London. For regression-based t-tests, summary(lm(y ~ x)) provides t-values and p-values for each regression coefficient automatically. Residual analysis is a natural companion skill when working with t-tests in regression contexts.

Python (SciPy and statsmodels)

Python’s scipy.stats library provides ttest_1samp(), ttest_ind(), and ttest_rel() for one-sample, independent two-sample, and paired t-tests respectively. Each function returns the t-statistic and p-value. For critical values: scipy.stats.t.ppf(0.975, df=20) returns the two-tailed critical value for df = 20 at α = 0.05. The statsmodels library offers more comprehensive output including confidence intervals through CompareMeans and DescrStatsW. Python is increasingly used in data science programs at universities and is the preferred tool in industry analytics roles. The combination of Pandas for data manipulation and SciPy for statistical tests makes Python a complete environment for t-test analysis without ever needing to open a printed t-distribution table.

Frequently Asked Questions About the T-Distribution Table