Normal Distribution Probability Calculator
Calculate the probability that a normally distributed random variable falls within a specified range, below a threshold, or above one — using the mean and standard deviation that fully describe the bell curve. Used for Six Sigma quality control, finance (VaR), psychometrics, and any process modelled by the normal distribution.
About this calculator
The normal (Gaussian) distribution is a symmetric bell-shaped probability distribution fully described by two parameters: its mean μ (centre) and standard deviation σ (spread). Probabilities are computed as areas under the curve. For a value of interest, standardise it to a z-score: z = (X − μ) / σ. The probability that X is between L and U is then P(L ≤ X ≤ U) = Φ(z_U) − Φ(z_L), where Φ is the cumulative distribution function (CDF) of the standard normal. Because Φ has no closed-form elementary expression, it is computed numerically via the error function: Φ(z) = 0.5 · [1 + erf(z / √2)]. This calculator evaluates that expression with the Abramowitz-Stegun rational approximation (accurate to ~7 decimal places), so you do not need a z-table. For tail probabilities: P(X < a) = Φ((a − μ)/σ); P(X > b) = 1 − Φ((b − μ)/σ). Key landmarks (the 68-95-99.7 rule): approximately 68% of probability lies within ±1σ of the mean, 95% within ±2σ, and 99.7% within ±3σ. Edge cases: σ must be > 0; if L = U the probability is 0 (a continuous distribution assigns zero probability to any single point); if L > U the calculator may return a negative result — swap them; very extreme tail probabilities (|z| > 5) are returned as ~0 because the rational approximation loses precision there, but the true probabilities are vanishingly small (P(|Z| > 5) ≈ 5.7 × 10⁻⁷). The normal model is only correct for variables whose generating process matches its assumptions (sum of many small independent additive effects). For skewed data (incomes, response times), heavy-tailed data (stock returns in a crisis), or bounded data (proportions), other distributions are appropriate.
How to use
Example 1 — Adult male height range. Adult male heights are approximately normal with μ = 175 cm, σ = 7 cm. What fraction of men are between 168 cm and 189 cm tall? Enter Mean = 175, Standard Deviation = 7, Lower Bound = 168, Upper Bound = 189, Type = between. z_L = (168 − 175)/7 = −1.00; z_U = (189 − 175)/7 = +2.00. P = Φ(2.00) − Φ(−1.00) ≈ 0.9772 − 0.1587 = 0.8185. ✓ About 81.9% of adult males fall in that height range. Example 2 — IQ above 130. IQ scores are normalised to μ = 100, σ = 15. What fraction of people have IQ above 130? Enter Mean = 100, Standard Deviation = 15, Lower Bound = 130, Upper Bound = 9999, Type = between (or use the dedicated "greater" option with Lower Bound = 130). z = (130 − 100)/15 = 2.00. P(X > 130) = 1 − Φ(2.00) ≈ 1 − 0.9772 = 0.0228. ✓ Approximately 2.28% of people have IQ ≥ 130 — the conventional "gifted" cutoff.
Frequently asked questions
What is the 68-95-99.7 rule and where does it come from?
For any normal distribution, approximately 68.27% of values lie within ±1 standard deviation of the mean, 95.45% within ±2 SDs, and 99.73% within ±3 SDs. These percentages come directly from Φ(1) ≈ 0.8413, Φ(2) ≈ 0.9772, and Φ(3) ≈ 0.99865, and the corresponding symmetric two-tail probabilities. The rule is extraordinarily useful for mental sanity checks: if a normally distributed quality measurement comes in at 3 SDs from the mean, that should happen on roughly 3 in 1,000 production units — so seeing it once is plausible, seeing it five times this morning is a red flag worth investigating. The rule is also the basis of "Six Sigma" quality programmes, which target processes where defects are 6+ standard deviations from the mean — a frequency of about 3.4 per million opportunities.
How do I find a one-tailed probability?
For P(X < a) (left tail), use Type = "less" and enter the value as a; the calculator returns Φ((a − μ)/σ). For P(X > b) (right tail), use Type = "greater" with the value as b; the calculator returns 1 − Φ((b − μ)/σ). If the calculator only supports between-bounds, simulate a left tail by setting Lower Bound to a very negative number (−9999) and Upper Bound to a, and simulate a right tail by setting Lower Bound to b and Upper Bound to +9999. The symmetry of the normal distribution means P(X < μ − k·σ) = P(X > μ + k·σ), so you can also exploit symmetry to convert between left and right tail problems without re-entering numbers.
When is the normal distribution actually appropriate for my data?
The normal distribution is appropriate when your data arises from the sum of many small independent additive effects — heights, measurement errors, IQ scores, manufacturing tolerances, and many biological traits naturally approximate it. The Central Limit Theorem also guarantees that the sample mean of almost any distribution becomes approximately normal for large enough samples, which is why so much of inferential statistics relies on the normal model. But: count data, times-to-event, proportions near 0 or 1, financial returns in periods of crisis, and heavily skewed variables (incomes, house prices, file sizes) are usually not normal, and applying normal probability calculations to them produces wrong answers — typically dramatic underestimates of tail probabilities. Always plot a histogram or Q-Q plot first; if the data shows obvious skew, multiple modes, or heavy tails, use a different distribution.
What are the most common mistakes people make with normal probability calculations?
The first is assuming normality without checking — financial returns are the classic example, where the normal assumption catastrophically understates tail risk (the 2008 crisis featured several "25-sigma" days that would essentially never happen in a true normal world). The second is mixing up cumulative and density: the normal density f(x) is not a probability and can take values above 1 for small σ — only the area under the curve is a probability. The third is forgetting that for continuous distributions, P(X = a) = 0 for any specific a; questions like "probability of exactly 175 cm" need to be reframed as a small interval around 175. The fourth is using sample mean and sample standard deviation as if they were the true population parameters in small samples; the resulting probabilities are slightly off and should be computed with a t-distribution for rigorous work. Finally, people sometimes mistake percentiles for probabilities — the 95th percentile is the value below which 95% of the data lies, not a probability itself.
When should I not use this calculator?
Skip it when the underlying variable is clearly non-normal — skewed (incomes, response times), discrete (count data, dice rolls), bounded (proportions, percentages), or heavy-tailed (financial returns in stress periods). For those, use the lognormal, Poisson, binomial, beta, or Student-t distributions respectively, each with its own probability calculator. Do not use it for very small probability events (z > 5) where the rational approximation degrades — for survival analysis, reliability engineering, or extreme-value statistics, use dedicated tail-probability tools. It is the wrong tool for joint or conditional probabilities of multiple normal variables; those require multivariate normal calculations and a covariance matrix. Finally, do not use it without first checking the normality assumption — a quick histogram or Q-Q plot is always worth the 30 seconds it takes.