Binomial Distribution Calculator: Exact, At Least & At Most

A seed packet says 80% of the seeds will sprout. You plant 20. How likely is it that at least 15 come up? The answer is 0.8042, or about 80.4%. That leaves roughly one packet in five giving you 14 or fewer seedlings, even when the 80% claim is completely true.

This calculator works out binomial probabilities for exactly, at most, at least, fewer than, more than or between any counts of successes. Type p as a decimal, a percent or a fraction. You get the full probability table, a bar chart with the normal curve on top, the mean and standard deviation, the most likely count, the normal approximation and how far off it is, and every step. A second tab answers the reverse question: how many trials you need for a given chance of success. It is set to the seed packet.

Binomial Distribution Calculator

p can be a decimal (0.25), a percent (25%) or a fraction (1/4). “Between” includes both end counts.

Binomial probability –
The distribution
Step by step
    Probability table

    Four Conditions, Checked on One Packet of Seeds

    The binomial distribution counts successes in a fixed number of tries. It applies only when all four of these hold:

    ConditionFor the seedsWhat would break it
    A fixed number of trials, n20 seeds, decided before plantingPlanting until 15 come up (that counts trials, not successes)
    Two outcomes per trialSprouts or doesn’tRecording “sprouted”, “sprouted late” and “didn’t” as three outcomes
    The same chance p every time80% for every seedMixing seeds from an old packet and a new one
    Independent trialsOne seed’s fate doesn’t change another’sPlanting them in a clump where one patch of soil floods

    When the conditions hold, the number of successes X has a binomial distribution with parameters n and p, written X ~ B(20, 0.8). If one fails, a different model fits better; the last section names them.

    Working Out P(at least 15) by Hand

    Every binomial probability comes from one formula:

    P(X = k) = C(n, k) × p<sup>k</sup> × (1 − p)<sup>n − k</sup>

    C(n, k) counts how many orders the k successes can come in. For exactly 15 sprouts out of 20:

    1. C(20, 15) = 15,504 different sets of 15 seeds that could be the ones that sprout.
    2. 0.8<sup>15</sup> = 0.035184, the chance that 15 particular seeds all sprout.
    3. 0.2<sup>5</sup> = 0.00032, the chance that the other 5 all fail.
    4. 15,504 × 0.035184 × 0.00032 = 0.1746.

    “At least 15” means 15, 16, 17, 18, 19 or 20, so add six terms:

    Sprouts k151617181920
    P(X = k)0.17460.21820.20540.13690.05760.0115

    The total is 0.8042. When the range you want is long, the shorter route is 1 minus the other side: P(X ≥ 15) = 1 − P(X ≤ 14) = 1 − 0.1958 = 0.8042. The calculator picks the shorter route on its own. When you type p as a fraction and the answer has a short enough fraction, it shows that too: with p = 1/3 and 6 trials, P(X = 2) = 80/243.

    At Least, More Than, At Most, Fewer Than: One Word Moves the Answer

    Counts are whole numbers, so “more than 15” starts at 16, while “at least 15” starts at 15. The difference is exactly P(X = 15), which here is 17.5 percentage points:

    QuestionIn symbolsCounts includedProbability
    At least 15P(X ≥ 15)15 to 200.8042
    More than 15P(X > 15)16 to 200.6296
    At most 15P(X ≤ 15)0 to 150.3704
    Fewer than 15P(X < 15)0 to 140.1958

    The complement rule needs the same care. P(X ≥ 15) = 1 − P(X ≤ 14), not 1 − P(X ≤ 15). Using 15 gives 0.6296, which is the answer to “more than 15”, a different question. This off-by-one slip turns up in printed tables and online calculators alike, so it pays to write out which counts the question includes before reaching for a formula.

    Reading a Range Off the Probability Table

    The table under the calculator lists, for every count k, the chance of exactly k, of k or fewer, and of k or more. Any range comes from two of those cumulative numbers. For the chance of 13 to 17 sprouts, take “17 or fewer” and remove “12 or fewer”:

    P(13 ≤ X ≤ 17) = P(X ≤ 17) − P(X ≤ 12) = 0.7939 − 0.0321 = 0.7618.

    The lower boundary is 12, not 13, because 13 belongs inside the range. The same answer comes from the other column: P(X ≥ 13) − P(X ≥ 18). Choosing “between” in the calculator does the subtraction for you and shades the bars it counted, which is a quick way to see whether the range you typed is the range you meant.

    Mean, Spread and the Most Likely Count

    The mean of a binomial distribution is n × p and the standard deviation is √(n × p × (1 − p)). For the seeds:

    • Mean: 20 × 0.8 = 16 sprouts.
    • Variance: 20 × 0.8 × 0.2 = 3.2, so the standard deviation is √3.2 = 1.789.
    • Within one SD (14.2 to 17.8, so 15, 16 or 17 sprouts): 59.8% of packets.
    • Within two SDs (12.4 to 19.6, so 13 to 19 sprouts): 95.6% of packets.

    The most likely single count is 16, yet it happens only 21.8% of the time. All 20 sprouting happens in 1.15% of packets. A packet with exactly the advertised 80% result is the exception, not the rule. The most likely count is found by rounding (n + 1) × p down; when (n + 1) × p is a whole number, that value and the one below it tie.

    How the Shape Changes With p

    At p = 0.5 the bars are symmetric around n/2. As p moves toward 0 the distribution leans right, with a long tail of rare high counts; toward 1 it mirrors that and leans left. The skewness is (1 − 2p) ÷ √(n·p·(1 − p)). For 20 trials it is +0.60 at p = 0.1, 0 at p = 0.5 and −0.60 at p = 0.9.

    The seeds sit at p = 0.8, so their bars pile up near 20 and trail off toward the low counts. Adding more trials reduces the lean, because the denominator grows with √n. That is the reason the normal curve fits better as n grows.

    The Normal Approximation and Its Half-Point Correction

    For large n, the bell curve with the same mean and SD gives nearly the same answers, and before computers it was the only practical method. A binomial count of k stands for everything from k − 0.5 to k + 0.5 on the continuous scale, so the boundaries move by half a unit. That shift is the continuity correction.

    Suppose 400 calls come into a support desk and each has a 30% chance of needing a callback. What is the chance of at most 130 callbacks? The mean is 120 and the SD is √84 = 9.165.

    • Exact binomial: 0.8736.
    • Normal with the correction: z = (130.5 − 120) ÷ 9.165 = 1.146, giving 0.8740.
    • Normal without it: z = (130 − 120) ÷ 9.165 = 1.091, giving 0.8624, more than a percentage point low.

    One rule of thumb asks for n·p and n·(1 − p) both at least 10; some textbooks use 5. Here they are 120 and 280, so the approximation is excellent. The seeds fail the rule, with n·(1 − p) = 4. The approximation still gives 0.7991 for “at least 15” against the exact 0.8042, but for “all 20 sprout” it gives 0.0252 against the exact 0.0115, more than double. The worst errors are in the tails, and the tails are exactly what a risk question asks about. The calculator shows the approximation, how far off it is and whether the rule is met, so you can see this for any n and p.

    Rare Events: When Poisson Beats the Normal Curve

    When p is small and n is large, the counts bunch up near zero and the bell curve fits badly. The Poisson distribution with λ = n × p does better.

    A factory knows 1 chip in 1,000 is faulty. In a batch of 2,000, what is the chance of 5 or more faulty chips? The mean is 2.

    MethodP(X ≥ 5)
    Exact binomial0.05256
    Poisson, λ = 20.05265
    Normal with correction0.0385

    The Poisson answer is off by about 0.0001; the normal one understates the risk by more than a quarter. The exact value needs no approximation, but knowing which one fits helps you check a printed answer.

    How Many Trials Until You Can Count On a Success?

    The Trials needed tab turns the question around. Suppose each job application has a 4% chance of leading to an interview, independently of the others.

    • On average the first interview comes after 1 ÷ 0.04 = 25 applications.
    • For a 50% chance of at least one interview you need 17 applications.
    • For a 95% chance you need 74, because 1 − 0.96<sup>74</sup> = 0.9512 while 73 gives only 0.9492.
    • For three interviews, the numbers rise to 67 applications for a 50% chance and 156 for 95%.

    For one success the formula is n ≥ ln(1 − target) ÷ ln(1 − p), rounded up. For more than one, the calculator searches for the smallest n where P(X ≥ r) reaches the target. The gap between the average (25) and the “almost sure” number (74) is the point: planning on the average leaves you short a large share of the time.

    Is This Result Unusual? A One-Sided Binomial Test

    A binomial probability can also act as a p-value. A supplier promises that at most 2% of parts are defective. You inspect 500 and find 20 defective, where 10 would be expected. If the promise were true, how often would 20 or more turn up?

    P(X ≥ 20) with n = 500 and p = 0.02 is 0.0031, about 1 in 320. That is strong evidence the true defect rate is above 2%. Had you found 15 defective, P(X ≥ 15) = 0.0814, which a careful buyer would note but not treat as proof.

    This is an exact binomial test. It avoids the normal approximation entirely, which matters with rare events like defects. For a test statistic that is already a z, t, χ² or F value, the p-value calculator gives the tail area instead.

    Binomial Probabilities in Excel and Google Sheets

    Both spreadsheets have the same functions:

    What you wantFormulaSeeds result
    Exactly 15=BINOM.DIST(15, 20, 0.8, FALSE)0.1746
    At most 15=BINOM.DIST(15, 20, 0.8, TRUE)0.3704
    At least 15=1 - BINOM.DIST(14, 20, 0.8, TRUE)0.8042
    Between 15 and 20=BINOM.DIST.RANGE(20, 0.8, 15, 20)0.8042
    Smallest k with P(X ≤ k) ≥ 5%=BINOM.INV(20, 0.8, 0.05)13

    The TRUE/FALSE switch decides between “at most” and “exactly”, and it is easy to leave on the wrong setting. BINOM.DIST.RANGE avoids the complement step: with one count it gives “exactly”, with two it gives the inclusive range. Microsoft’s documentation shows it with 60 trials at 75%, where exactly 48 successes comes to 0.084 and 45 to 50 successes to 0.524; the calculator gives the same. BINOM.INV here returns 13 because P(X ≤ 12) = 0.032 is below 5% and P(X ≤ 13) = 0.087 is not.

    When the Binomial Is the Wrong Model

    Many problems look binomial and aren’t. Two cases show how much the answer can move.

    Drawing without replacement from a small group. Pick 4 people at random from a group of 10, of whom 6 are women. Each pick changes the chances for the next, so the trials aren’t independent. The correct (hypergeometric) chance that all 4 are women is 0.0714; the binomial with p = 0.6 gives 0.1296, nearly double. For exactly 2 women it is 0.4286 against the binomial’s 0.3456. From a group of 1,000 with 600 women, the true answer for all 4 is 0.1291, almost the binomial value. A rule of thumb (the 10% condition) treats the binomial as close enough when the sample is under a tenth of the group.

    A different chance on each trial. Three exams with pass chances of 90%, 70% and 50% average 70%, but treating them as three 70% trials is wrong. The chance of passing all three is 0.9 × 0.7 × 0.5 = 0.315, not 0.7<sup>3</sup> = 0.343, and the chance of failing all three is 0.015, not 0.027.

    SituationModel to use
    Fixed n, same p, independent trialsBinomial (this calculator)
    Sampling without replacement from a small groupHypergeometric
    Each trial has its own chancePoisson binomial
    Counting trials until the first (or r-th) successGeometric (negative binomial)
    Events in time or space with no fixed nPoisson
    More than two outcomes per trialMultinomial

    The Trials needed tab handles the “until the r-th success” question from the binomial side, by finding the n that makes r successes likely enough. For questions about which items get picked rather than how many succeed, the permutation and combination calculator counts the possibilities directly.