Study Tips Hypothesis Testing 11 min read Updated November 14, 2026
BY: Minsa A (Senior Statistics Editor)
Reviewed By: Statistics Fundamentals Team

T Score vs Z Score: When to Use Each (With Examples)

Two test statistics, one decision. A t score and z score both measure how far a sample mean sits from a hypothesized population mean — but they draw on different information. Getting the choice wrong gives you p-values you can't trust.

Z Score ↓ T Score ↓ Comparison Table ↓ Decision Rule ↓ Convergence ↓ Worked Examples ↓ Bone Density ↓ FAQ ↓
Quick Answer — Featured Snippet

A t score and z score are both standardized test statistics, but they differ in one key way: a z score requires a known population standard deviation (σ), while a t score uses the estimated sample standard deviation (s). Use a z score when σ is known and your sample is large (n ≥ 30). Use a t score when σ is unknown — which covers most real-world research.

What Is a Z Score?

A z score (also called a z-statistic or z-test statistic) measures how many standard deviations a value lies from the population mean. It sits on the standard normal distribution, which has a mean of 0 and a standard deviation of 1. The key requirement: you must already know the population standard deviation σ.

There are two versions of the formula depending on what you are testing. Many sources show only one, which produces a critical calculation error:

Z Score — Two Forms
For a single observation: z = (x − μ) / σ
For a sample mean: z = (x̄ − μ) / (σ / √n)
x̄ = sample mean μ = population mean σ = population standard deviation (must be known) σ/√n = standard error of the mean
⚠️
Common formula error

When testing a sample mean, the denominator is σ/√n (the standard error), not σ alone. Using σ directly inflates the z-score by a factor of √n — a fivefold error when n = 25. Always divide by the standard error when x̄ is your test value.

The z-score result is compared against standard normal critical values: ±1.96 for a two-tailed test at α = 0.05, or ±1.645 for a one-tailed test at the same alpha. Use the full z score calculator for quick results.

What Is a T Score?

A t score (t-statistic) is used when σ is unknown and must be estimated from the data as the sample standard deviation s. Because s varies from one sample to the next, the t-distribution has heavier tails than the standard normal — it accounts for the extra uncertainty in your denominator. This makes t-tests more conservative: they require a larger test statistic to claim significance.

T Score Formula
t = (x̄ − μ) / (s / √n)
x̄ = sample mean μ = hypothesized population mean s = sample standard deviation (estimated from data) n = sample size df = n − 1 degrees of freedom

The t score is compared against t-distribution critical values that depend on degrees of freedom (df = n − 1). As df grows toward infinity, the t-distribution converges to the standard normal.

The t-distribution was developed by William Sealy Gosset in 1908, publishing under the pen name "Student" because Guinness Brewery prohibited employees from publishing under their own names. Gosset needed a reliable method to test small batches of barley with no access to population parameters — a situation that describes most modern research just as well.

For hands-on calculations, the one-sample t-test guide walks through the full testing procedure.

T Score vs Z Score: Side-by-Side Comparison

This table is in HTML text — not an image. All seven pages currently ranking for this keyword use either an image-based table or no table at all, meaning their comparison data is invisible to search engines and screen readers.

FeatureZ ScoreT Score
Population SD (σ) required ✓ Yes — must be known before the test ✗ No — estimated as sample SD (s)
When σ is known ✓ Use z ✗ Using t is valid but less efficient
When σ is unknown ✗ Do not use z — anti-conservative p-values ✓ Always use t
Distribution used Standard normal (z) Student's t-distribution
Degrees of freedom Not needed df = n − 1
Critical value at α=0.05 (two-tailed) ±1.960 Varies: 2.228 at df=10; 2.045 at df=29
Accuracy for small samples Underestimates uncertainty without known σ ✓ More accurate — heavier tails protect the test
Typical real-world use Rare — σ is almost never known ✓ The standard choice in most research

The One Rule That Decides Everything

Several sources give conflicting rules — one popular Q&A site lists three different rules across three tutors, all contradicting each other. The correct rule has two steps:

1
Is the population standard deviation (σ) known?

If yes — from prior research, official records, or engineering specifications — use a z score. This situation is rare in practice. Examples: IQ tests (σ = 15, by design), standardized manufacturing tolerances. If σ is unknown, go to Step 2.

2
Is σ unknown? Use a t score — regardless of sample size.

When σ is unknown, the t-test is always correct. Sample size affects how close t is to z in practice, but it does not change which test to use. With n < 30, the t-distribution's heavier tails are essential. With n ≥ 30, t and z give nearly the same result — but t remains technically more accurate because σ is still estimated.

σ unknown → t-test, regardless of n.   σ unknown + large n → still t-test.   Using t is never the wrong choice when σ is estimated.
⚠️
The "n ≥ 30" shortcut is not a rule

Some textbooks say "use z when n ≥ 30." This works because t ≈ z for large samples, but it is a shortcut, not a statistical principle. The σ-known requirement for z has nothing to do with sample size. If σ is unknown and n = 500, the t-test is still more precise.

In medical studies, psychology experiments, business analytics, and social research, σ is virtually never known in advance. That means the t-test is the standard tool for the overwhelming majority of real-world hypothesis testing.

Why T and Z Scores Converge for Large Samples

As sample size grows, the sample standard deviation s becomes a more precise estimate of σ. The extra uncertainty that makes t-distribution tails heavier gradually shrinks. At df = ∞ (an infinitely large sample), the t-distribution is identical to the standard normal.

One commonly cited figure is that "t is within 3% of z once df exceeds 29." That figure is wrong. The table below shows the actual values:

Critical Value Convergence Table (Two-Tailed, α = 0.05)

Sample size (n) Degrees of freedom (df) t critical value z critical value Gap above z
652.5711.96031.2%
11102.2281.96013.7%
21202.0861.9606.4%
30292.0451.9604.3% ← n=30 is NOT "within 3%"
41402.0211.9603.1%
61602.0001.9602.0%
1211201.9801.9601.0%
∞∞1.9601.9600.0%
ℹ️
What this table shows

At n = 30 (df = 29), the t critical value is 2.045 — still 4.3% above z = 1.960. The gap only falls below 3% around df = 40 (n = 41), and below 2% around df = 60 (n = 61). The 3% threshold is not crossed at the boundary most textbooks use. Use the critical value calculator to find any df-specific value.

Worked Examples

Example 1 — When to Use a Z Score (σ Known)

Example 1 — Z Score

Testing whether a cohort differs from a known population (σ known)

A standardised aptitude test has been administered for decades and has a known population mean of μ = 500 and population standard deviation of σ = 100. A new cohort of 36 students produces a sample mean of x̄ = 520. Is this cohort significantly different from the population?

1

σ is known (σ = 100). Use the z-test.

2

Calculate the standard error: σ/√n = 100/√36 = 100/6 ≈ 16.67
Note: use the standard error (σ/√n), not σ alone — a critical distinction when testing sample means.

3
z = (x̄ − μ) / (σ/√n) = (520 − 500) / 16.67 = 20 / 16.67 = 1.20
4

Critical value at α = 0.05 (two-tailed) = ±1.96. Since |z| = 1.20 < 1.96, we fail to reject H₀.

✓ z = 1.20, p ≈ .230. No statistically significant difference from the population. Use the z score calculator to verify.

Example 2 — When to Use a T Score (σ Unknown)

Example 2 — T Score

Testing a sleep intervention on reaction time (σ unknown)

A researcher tests whether a sleep intervention improves reaction time. A sample of n = 15 participants shows a mean improvement of x̄ = 12 ms with a sample standard deviation of s = 18 ms. The null hypothesis is that there is no change (μ = 0).

1

σ is unknown — it was not measured before the study. Use the t-test.

2

Degrees of freedom: df = n − 1 = 14

3
t = (x̄ − μ) / (s/√n) = (12 − 0) / (18/√15) = 12 / (18/3.873) = 12 / 4.648 = 2.58
4

Critical value at α = 0.05, df = 14 (two-tailed) = ±2.145. Since t = 2.58 > 2.145, we reject H₀.

✓ t(14) = 2.58, p = .022. The sleep intervention produced a statistically significant improvement. See the one-sample t-test guide for the full procedure.

ℹ️
What would happen if we used z incorrectly in Example 2?

With z = 1.96 as the critical value instead of t = 2.145, the same test statistic of 2.58 still clears the bar. But at smaller samples or borderline test statistics, the difference matters. At df = 5 with t = 2.57, the z critical of 1.96 would call significance where t correctly requires 2.571 — meaning z would falsely reject H₀.

Bone Density T-Score vs Statistical T-Score: They Share a Name, Nothing Else

If you arrived here from a search about a DXA or DEXA scan result, the statistical content on this page does not apply to your situation. The bone density T-score and the statistical t-score share naming conventions only — they measure entirely different things.

FeatureStatistical T-ScoreBone Density T-Score
What it measuresTest statistic for hypothesis testingBone mineral density vs young adult mean
FieldStatistics and research methodologyOrthopaedics and osteoporosis medicine
Reference groupHypothesized population meanHealthy 30-year-old adults (peak bone mass)
Formulat = (x̄ − μ) / (s/√n)(Patient BMD − Young adult mean BMD) / Young adult SD
InterpretationCompare to t-critical value at chosen αSee WHO classification below
Also uses "Z-score"?Yes — the z-test for known σYes — but means age-matched peers, not young adults

WHO Bone Density T-Score Classification

T-ScoreDiagnosisWhat it means
≥ −1.0Normal bone densityBone density within expected range for peak mass
−1.0 to −2.5OsteopeniaLow bone density — monitoring recommended
≤ −2.5OsteoporosisSignificantly reduced bone density — treatment likely needed

The bone density Z-score (not T-score) compares a patient's bone density to age- and sex-matched peers rather than young adults. A bone density Z-score ≤ −2.0 suggests bone loss exceeding what is expected for the patient's age group. For clinical interpretation of scan results, consult your physician or rheumatologist.

Five Mistakes to Avoid

1
Using z when σ is unknown

Without knowing σ, you are substituting s into a formula that assumes the denominator is a fixed population parameter. The result is anti-conservative p-values — significance is too easy to reach, and your Type I error rate exceeds the stated alpha.

2
Using σ directly when testing a sample mean

The z formula for a sample mean is z = (x̄ − μ) / (σ/√n). Using σ alone in the denominator treats your sample mean as though it were a single individual observation — producing a z-score that is off by a factor of √n.

3
Treating "n ≥ 30" as permission to use z

The n ≥ 30 guideline describes when t ≈ z in practice. It does not change the requirement that σ must be known to justify a z-test. When σ is estimated from the sample, the t-test remains the correct tool regardless of sample size.

4
Confusing bone density T-scores with statistical t-scores

A DXA scan T-score of −2.5 does not mean "2.5 standard deviations below significance." It means bone density 2.5 standard deviations below the mean for a young adult reference population. The clinical and statistical uses of "T-score" are unrelated — see the section above.

5
Forgetting degrees of freedom

The t critical value is not a fixed number. At df = 10 it is 2.228; at df = 29 it is 2.045; at df = ∞ it is 1.960. Using the wrong df — or the z critical value by mistake — produces an incorrect significance threshold. Always look up or calculate the t critical for your specific df. Use the critical value calculator to avoid this.

APA Reporting Format

When reporting statistical results, the format matters as much as the numbers. Here are the standard APA 7th edition formats for z and t results:

Z result z = 1.20, p = .230
T result t(14) = 2.58, p = .022

The number in parentheses after t is the degrees of freedom. Always report p-values alongside the test statistic — do not report only "p < .05." State whether the test was one-tailed or two-tailed in your methods section. For t-tests, consider reporting Cohen's d to convey practical significance alongside statistical significance. The p-value calculator converts test statistics to exact p-values.

Frequently Asked Questions

What is the main difference between a t score and a z score? ▼

A z score uses the known population standard deviation (σ), while a t score uses the estimated sample standard deviation (s). Because s varies from sample to sample, the t-distribution has heavier tails than the standard normal, especially for small samples. As sample size grows, the t-distribution converges to the standard normal (z) distribution.

When should I use a t score instead of a z score? ▼

Use a t score whenever the population standard deviation (σ) is unknown, which covers the vast majority of real-world research. The only time you use a z score is when σ is known from prior data, census records, or engineering specifications — a rare condition in practice.

Can I use a z score for small samples? ▼

Only if the population standard deviation σ is known. If σ is unknown and your sample is small (n < 30), you must use a t score. Using a z score without knowing σ gives incorrect p-values because it underestimates the uncertainty in your standard deviation estimate.

Do t scores and z scores ever give the same result? ▼

They converge as sample size increases. At n = 30 (df = 29), the t critical value is 2.045 versus 1.960 for z — a 4.3% gap. By df = 120, the gap falls to under 1%. Statisticians often treat them as interchangeable for n ≥ 30 when σ is unknown, though technically the t-test is always more accurate when using sample standard deviation.

What is a "good" t score or z score? ▼

Whether a t score or z score indicates significance depends on your chosen alpha level. For a two-tailed test at α = 0.05, the critical z value is ±1.96. The critical t value at the same alpha depends on degrees of freedom — at df = 10 it is ±2.228, at df = 30 it is ±2.042. Any score beyond the critical value indicates statistical significance.

What is the difference between a bone density T-score and a statistical t-score? ▼

They are completely different concepts that share the same name. A bone density T-score (from DXA/DEXA scans) measures how much your bone density differs from the mean for a healthy 30-year-old adult. A statistical t-score is a test statistic calculated from sample data to perform a hypothesis test. A bone density T-score of −2.5 or below indicates osteoporosis; a statistical t-score is compared to a t-distribution critical value to get a p-value.

What formula do I use for a z score of a sample mean? ▼

When testing whether a sample mean (x̄) differs from a population mean (μ), use z = (x̄ − μ) / (σ/√n), where σ/√n is the standard error of the mean. A common mistake is using σ directly instead of σ/√n, which gives an incorrect and inflated z-score result — the z-value will be off by a factor of √n.

What happens when σ is unknown but n is large? ▼

When σ is unknown and n is large (typically n ≥ 30), you should still use a t-test with the sample standard deviation (s). For large n, the t-distribution is nearly identical to the standard normal, so the results are virtually the same as a z-test. Some textbooks allow the z approximation here, but using the t-test is never wrong and is always more precise.

Key Takeaways

📌

σ known → z score. σ unknown → t score. This is the whole decision, not sample size.

📌

When testing a sample mean, the z denominator is σ/√n (the standard error), not σ alone. Using σ inflates the z-score by √n.

📌

σ unknown + large n → still use t. Results will be nearly identical to z, but t is technically correct.

📌

At n = 30 (df = 29), the t critical value is 2.045 — still 4.3% above z = 1.960. The gap only falls below 3% around n = 41.

📌

Bone density T-score ≠ statistical t-score. They share a name but measure entirely different things.

📌

T critical values depend on degrees of freedom (df = n − 1). The t-table is not the z-table.

References and Further Reading
Gosset, W. S. ("Student") — "The Probable Error of a Mean," Biometrika, 1908 | Freedman, Pisani & Purves — Statistics (4th ed.) | WHO Technical Report 843 — Assessment of Fracture Risk (bone density classification) | For related tools: t-distribution table, z table, statistics cheat sheet, types of statistical tests.