The empirical rule is one of the most practical shortcuts in applied statistics. Also called the 68-95-99.7 rule, it tells you exactly what proportion of a normally distributed dataset falls within one, two, and three standard deviations of the mean — without looking up a table or running a calculation. Empirical rule statistics gives you an instant mental model of data spread whenever the distribution is approximately bell-shaped.
The rule works because the normal distribution’s shape is mathematically fixed. Once you know the mean and standard deviation, the proportion of observations inside each symmetric interval is locked in:
- 68% of observations fall within 1 standard deviation of the mean
- 95% of observations fall within 2 standard deviations of the mean
- 99.7% of observations fall within 3 standard deviations of the mean
This article explains where those percentages come from, walks through a fully worked numeric example, and shows how to apply the rule to real-world data. Use it alongside the calculator at the bottom to move beyond the three-point approximation and find exact probabilities for any interval.
What Is the Empirical Rule?
The empirical rule describes the spread of data in any normal distribution — a symmetric, bell-shaped probability distribution where most values cluster near the center and frequencies taper off evenly toward both tails. The rule was established from direct observation of how normally distributed measurements behave, which is why it is called “empirical” (derived from observation) rather than “theoretical.”
The three intervals are symmetric around the mean. If the mean is μ and the standard deviation is σ, then:
68% of data lies in: μ − 1σ to μ + 1σ
95% of data lies in: μ − 2σ to μ + 2σ
99.7% of data lies in: μ − 3σ to μ + 3σ
The remaining 0.3% of data lies beyond three standard deviations on both tails — genuinely rare values.
This rule is the practical entry point for dozens of statistical methods and quality-control procedures. In manufacturing, a process running at “three sigma” guarantees that 99.7% of products meet specification. In biomedical research, reference ranges for lab tests are typically set as “mean ± 2 standard deviations,” capturing the middle 95% of healthy patients. In finance, events beyond two standard deviations from the mean are flagged as statistically unusual.
The Three Intervals in Detail
68%: Within One Standard Deviation
Roughly two-thirds of all observations in a normal distribution fall within one standard deviation of the mean. If a population has mean μ = 100 and σ = 15, then about 68% of values lie between 85 and 115.
The remaining 32% splits equally across both tails: approximately 16% below 85 and 16% above 115. This interval is the core of much practical statistics work. When a lab analyst reports that a measurement is “within one sigma of the reference value,” they are saying the probability of landing that far from the center purely by chance is about 68%.
95%: Within Two Standard Deviations
About 19 out of every 20 observations fall within two standard deviations of the mean. Using the same example (μ = 100, σ = 15), 95% of values lie between 70 and 130.
The remaining 5% splits symmetrically: about 2.5% below 70 and 2.5% above 130. This two-sigma boundary is the direct origin of the most common significance threshold in statistics. When researchers set α = 0.05, they are designating the outermost 5% of the normal curve as the rejection region — exactly the portion outside the two-sigma interval. The critical z-score of ±1.96 is just slightly inside the two-sigma mark (1.96 vs. 2.00) because the precise area from −1.96 to +1.96 equals 95.00%, not 95.45%.
99.7%: Within Three Standard Deviations
Three standard deviations from the mean capture almost all observations — 99.7%, or 997 out of 1000. Values beyond this boundary are exceptional. With μ = 100 and σ = 15, fewer than 3 people per 1000 would score below 55 or above 145.
In finance and risk management, a “five-sigma event” is theoretically so improbable under a normal model that it should almost never occur in a human lifetime — yet extreme market crashes happen far more often than the model predicts. This mismatch reveals that financial returns have heavier tails than the normal distribution, and that blindly applying the empirical rule to non-normal data is dangerous.
Why These Specific Percentages?
The 68%, 95%, and 99.7% values come directly from the mathematics of the standard normal distribution. The area under the bell curve between −1 and +1 standard deviation (in a distribution with mean 0 and standard deviation 1) evaluates to exactly 68.27%. These areas are tabulated in every z-table and built into every statistics software package.
The more precise values are:
| Interval | Exact percentage | Rounded |
|---|---|---|
| μ ± 1σ | 68.27% | 68% |
| μ ± 2σ | 95.45% | 95% |
| μ ± 3σ | 99.73% | 99.7% |
The compact “68-95-99.7” form is the practical shorthand. For most teaching and applied uses, the rounding is negligible. For exact hypothesis tests or confidence-interval calculations, use the z-table or calculator directly.
According to the NIST/SEMATECH e-Handbook of Statistical Methods — Normal Distribution, the normal distribution is entirely described by its mean and standard deviation, and all probability statements about it reduce to areas under the standardized bell curve — areas that the empirical rule summarizes at the three most useful thresholds.
Fully Worked Example: Heights of Adult Women
Suppose adult women’s heights in a population are approximately normally distributed with:
- Mean (μ) = 165 cm
- Standard deviation (σ) = 7 cm
Apply the empirical rule step by step.
Step 1: Compute the One-Sigma Interval
Lower bound: μ − 1σ = 165 − 7 = 158 cm
Upper bound: μ + 1σ = 165 + 7 = 172 cm
Result: About 68% of adult women in this population are between 158 cm and 172 cm tall.
Step 2: Compute the Two-Sigma Interval
Lower bound: μ − 2σ = 165 − 14 = 151 cm
Upper bound: μ + 2σ = 165 + 14 = 179 cm
Result: About 95% of adult women are between 151 cm and 179 cm tall. Only 1 in 20 women is shorter than 151 cm or taller than 179 cm.
Step 3: Compute the Three-Sigma Interval
Lower bound: μ − 3σ = 165 − 21 = 144 cm
Upper bound: μ + 3σ = 165 + 21 = 186 cm
Result: About 99.7% of adult women fall between 144 cm and 186 cm tall. Fewer than 3 in 1000 women fall outside this range.
Step 4: Answer a Probability Question
Question: What proportion of women are taller than 179 cm?
179 cm is exactly two standard deviations above the mean (μ + 2σ). The 95% rule tells us 5% of data falls outside the two-sigma interval, split equally across both tails:
P(height > 179 cm) ≈ 5% ÷ 2 = 2.5%
About 1 in 40 women exceeds 179 cm in this population.
Question: What proportion are between 158 cm and 179 cm?
- 158 cm = μ − 1σ (the lower edge of the one-sigma band)
- 179 cm = μ + 2σ (the upper edge of the two-sigma band)
From the mean downward to 158 cm: half of the one-sigma band = 68% ÷ 2 = 34% From the mean upward to 179 cm: half of the two-sigma band = 95% ÷ 2 = 47.5%
P(158 < height < 179) ≈ 34% + 47.5% = 81.5%
Try the Calculator
The normal distribution calculator below lets you verify these numbers and explore any mean and standard deviation. Enter μ = 165, σ = 7, and a specific x value to find the exact probability — a more precise answer than the empirical rule’s three-point shorthand.
When the Empirical Rule Applies
The empirical rule is valid only for normal (bell-shaped) distributions. Before applying it, check that your data reasonably satisfies the normality assumption.
Signs the Rule Applies
A histogram of the data shows a symmetric, single-peaked, bell-shaped curve. The mean and median are approximately equal (a sign of symmetry). Skewness is close to zero. Sample sizes are large — thirty or more is a common rule of thumb. The variable arises from many small independent factors added together: heights, exam scores, blood pressure readings, measurement errors, manufacturing tolerances, and similar continuous measurements all tend to be approximately normal by the central limit theorem.
Signs the Rule Does Not Apply
Skewed distributions: Income, house prices, and reaction times are right-skewed. Applying the empirical rule would badly misestimate the proportion of low-income versus high-income earners.
Heavy-tailed distributions: Returns on financial assets have more extreme events than a normal model predicts. Using the empirical rule here systematically underestimates tail risk.
Bimodal distributions: Two peaks signal two mixed groups. The empirical rule is invalid without first separating the subpopulations.
Small or discrete datasets: A dataset of ten coin-flip outcomes is not normal. The rule requires a continuous measurement distribution with enough observations to approximate the bell curve.
When normality fails, Chebyshev’s theorem provides a safer, distribution-free alternative (discussed below).
Real-World Applications of the Empirical Rule
Quality Control and Six Sigma
Manufacturing processes use the empirical rule to define specification limits. A process running at “three sigma” places 99.7% of products within the acceptable range — a defect rate of about 3 per 1000. “Six sigma” extends this to six standard deviations from the mean, a defect rate of about 3.4 per million, achieved by also accounting for real-world process drift of ±1.5σ over time.
Quality engineers use the empirical rule to set control-chart warning lines (one and two sigma) and action lines (three sigma) on statistical process control charts. When a data point falls outside three sigma, it triggers an investigation — not because it is certainly a defect, but because the probability of such a point arising from a stable process is only 0.15% on each side.
Standardized Tests and Education
SAT, ACT, IQ, and many other standardized tests are designed to produce approximately normal score distributions with a published mean and standard deviation. Knowing the empirical rule lets students and educators immediately interpret percentile standing:
- A score one standard deviation above the mean sits at roughly the 84th percentile (the top 16%, since 16% of scores exceed the one-sigma upper bound)
- A score two standard deviations above the mean sits at roughly the 97.5th percentile
Medical Reference Ranges
Clinical laboratories publish reference ranges for continuous biomarkers — cholesterol, blood glucose, white-blood-cell counts — as “mean ± 2 standard deviations” for a healthy reference population. This captures the middle 95% of healthy patients and defines the boundary of “normal.” A result outside this range is flagged for clinical follow-up because it falls in the bottom 2.5% or top 2.5% of the reference distribution.
Grading on a Curve
When instructors “grade on a curve,” they often assign letter grades based on how many standard deviations above or below the class mean a student scores. The empirical rule immediately shows how many students would receive each grade under a pure normal model: about 68% score within one sigma of the mean (a B or C range in many schemes), about 27% more fall between one and two sigma on either side (A and D), and only about 4.6% fall between two and three sigma.
Empirical Rule vs. Chebyshev’s Theorem
A common point of confusion is when to use the empirical rule versus Chebyshev’s theorem (also written Chebyshev’s inequality).
Chebyshev’s theorem makes a weaker but universally applicable statement: for any distribution with a finite mean and standard deviation — regardless of shape — at least 1 − 1/k² of all observations fall within k standard deviations of the mean. Setting k = 2:
At least 1 − 1/4 = 75% of data lies within 2σ (any distribution)
Compare that to the empirical rule’s 95% — valid only when the distribution is normal.
| Theorem | Within 1σ | Within 2σ | Within 3σ | Applies to |
|---|---|---|---|---|
| Empirical rule | ≈ 68% | ≈ 95% | ≈ 99.7% | Normal distributions only |
| Chebyshev’s theorem | ≥ 0% | ≥ 75% | ≥ 88.9% | Any distribution |
Chebyshev’s theorem gives a floor — the true percentage could be much higher. The empirical rule gives the precise value, but only when normality holds.
When to use which:
- Confirmed normal distribution → empirical rule (tighter, more informative)
- Unknown or non-normal distribution → Chebyshev’s theorem (always valid)
- Large sample of a continuous measurement → test normality first, then decide
The OpenStax Introductory Statistics textbook — Chapter 2, Descriptive Statistics covers both the empirical rule and Chebyshev’s theorem in full, with additional worked examples and practice problems freely available online.
Common Mistakes When Using the Empirical Rule
Applying It to Non-Normal Data
The most frequent error is using the 68-95-99.7 rule on skewed or multimodal data. Always check a histogram for rough bell symmetry before applying the rule. When in doubt, use Chebyshev’s theorem — it is always valid.
Confusing Population and Sample Parameters
The empirical rule uses population parameters: μ (the population mean) and σ (the population standard deviation). In practice you usually work with a sample, so you substitute x̄ (the sample mean) and s (the sample standard deviation). The rule still applies when the sample is large enough to approximate a normal distribution, but x̄ and s are estimates — the realized percentages may differ slightly from 68%, 95%, 99.7%.
Misreading “Within 2 Standard Deviations”
“Within 2 standard deviations” means the full two-sided interval from μ − 2σ to μ + 2σ, covering 95% of data. A common error is reading it as only the distance from the mean to the upper bound (μ to μ + 2σ), which is only 47.5% of data. The rule always refers to a symmetric, two-sided interval.
Forgetting That the Intervals Are Approximate
The rule gives rounded percentages. The exact value for the one-sigma interval is 68.27%, not 68.00%. For most teaching and mental-math purposes this rounding is harmless. When calculating a hypothesis test critical value or a precise confidence interval, use the z-table or the normal distribution calculator directly — the approximation is a teaching tool, not a computational one.
Treating Outliers as Impossible
The empirical rule does not say that values beyond three sigma are impossible — it says they account for 0.3% of observations. In a large dataset of 10,000 observations, the empirical rule predicts about 30 values (0.3%) beyond three sigma. Treating every three-sigma data point as an error or fraud is a misapplication of the rule. It is unusual, not impossible.
Frequently Asked Questions
What does the empirical rule state?
The empirical rule states that in a normal distribution, approximately 68% of observations fall within one standard deviation of the mean, 95% within two standard deviations, and 99.7% within three standard deviations. These three thresholds provide a rapid description of where data concentrates in any bell-shaped distribution.
What is the empirical rule in statistics?
In statistics, the empirical rule (also called the 68-95-99.7 rule) is a description of how data in a normal distribution spreads around the mean. It allows statisticians to estimate what proportion of a dataset lies within a given number of standard deviations from the mean, without computing a full integral or consulting a complete probability table.
How do you apply the empirical rule step by step?
- Confirm the data is approximately normally distributed (check a histogram for a bell shape).
- Identify the mean (μ) and standard deviation (σ).
- Compute the one-sigma interval: μ − σ to μ + σ (approximately 68% of data).
- Compute the two-sigma interval: μ − 2σ to μ + 2σ (approximately 95% of data).
- Compute the three-sigma interval: μ − 3σ to μ + 3σ (approximately 99.7% of data).
- Use these bounds to answer probability questions or to define what counts as an outlier.
What does it mean that 95% falls within two standard deviations?
It means that if you select a value at random from a normally distributed population, there is a 95% probability that it falls between μ − 2σ and μ + 2σ. The remaining 5% is split equally across the two tails: 2.5% below the lower bound and 2.5% above the upper bound. This is the direct basis for the conventional α = 0.05 significance level in hypothesis testing.
Does the empirical rule work for skewed data?
No. The empirical rule requires a normal (symmetric, bell-shaped) distribution. For skewed data it produces incorrect estimates. Use Chebyshev’s theorem instead — it guarantees that at least 75% of data falls within two standard deviations and at least 88.9% within three standard deviations, regardless of the distribution’s shape.
How does the empirical rule relate to standard deviation?
Standard deviation measures how spread out a distribution is around its mean. The empirical rule translates that abstract spread into concrete proportions: one standard deviation captures 68% of data, two standard deviations capture 95%, three capture 99.7%. This is why standard deviation is such a powerful summary statistic — paired with the empirical rule, a single number (σ) immediately communicates the full spread of a normal distribution in terms a non-statistician can understand.
What is the difference between the empirical rule and the rule of large numbers?
These are unrelated concepts. The law of large numbers says that a sample mean converges toward the true population mean as the sample grows — a statement about estimation. The empirical rule describes how individual observations distribute around a fixed mean in a normal distribution — a statement about spread. Knowing one tells you nothing about the other.
Can I use the empirical rule to find outliers?
Yes, the three-sigma rule is a simple outlier criterion: any data point more than three standard deviations from the mean falls in the 0.3% tail and is often treated as a potential outlier. This criterion is built into many statistical process control methods. However, it is only valid when the data is approximately normal, and the 0.3% probability means genuine three-sigma values do occur — especially in large datasets — so always investigate context before discarding an extreme value.
Summary
The empirical rule (68-95-99.7 rule) is the most useful shorthand for understanding how data spreads around a normal distribution mean. The three thresholds — 68%, 95%, and 99.7% within one, two, and three standard deviations — appear throughout statistics, medicine, manufacturing, finance, and education precisely because so many real-world measurements follow an approximately normal distribution.
The key constraint is normality: empirical rule statistics only works when the underlying distribution is bell-shaped. When normality is uncertain or absent, Chebyshev’s theorem provides a safe, distribution-free alternative. When normality holds and you need more precision than the three-point approximation, use the normal distribution calculator to compute exact probabilities for any interval.
You can also explore all statistics tools at the calculators hub.