The Lady Tasting Tea: A Famous Statistics Test You Can Try With Coffee

Can you really tell your favorite coffee from another one, or does the label do half the work? A blind coffee tasting answers that question, but only if it is designed well. Luckily, statisticians solved this problem about a century ago, over a cup of tea.

The famous lady tasting tea experiment, described by the statistician Ronald Fisher, is one of the founding stories of modern statistics. Its logic works perfectly for coffee. Below: the story, the math (with worked examples), the triangle test used by professional tasters, and a simple plan for your own blind tasting.

Key takeaways

• In Fisher’s design, a taster sorts 8 cups (4 of each kind). Guessing gets all 8 right with a chance of only 1 in 70, about 1.4%.
• Getting 3 of 4 right sounds impressive, but happens by luck about 24% of the time.
• In a triangle test, you pick the odd one out of 3 cups; a guess is right 1 time in 3.
• With 6 triangle rounds, 5 or more correct would happen by chance less than 2% of the time.
• Randomize, hide labels and decide your pass mark before you start.

In this article

1. The story behind the lady tasting tea
2. The math: why 1 in 70?
3. Significance in plain words
4. The triangle test used in coffee
5. How to run a blind coffee tasting at home
6. FAQ

The story behind the lady tasting tea

In 1919, Ronald A. Fisher joined Rothamsted Experimental Station, an agricultural research center north of London. During an afternoon tea break there in the early 1920s, as the story goes, Fisher offered a cup of tea to a colleague, the algae scientist Dr. Muriel Bristol. She declined: he had poured the milk in first, and she preferred tea poured first. She claimed she could taste the difference.

Fisher was skeptical. A colleague, William Roach (who later married Bristol), suggested testing her on the spot. The accounts, including Fisher’s daughter Joan Fisher Box’s 1978 biography and later retellings, report that Bristol identified the cups correctly. The exact date was not recorded, so treat the details of that afternoon as well-attested anecdote rather than a documented experiment.

What is documented is what Fisher did with the idea. In his 1935 book The Design of Experiments, he used a tea-tasting lady as the opening example to explain how to design a fair test, and the method became a cornerstone of experimental science.

The math: why 1 in 70?

Fisher’s design is simple:

• Prepare 8 cups: 4 with milk poured first, 4 with tea poured first.
• Serve them in random order.
• Tell the taster there are exactly 4 of each, and ask her to pick out the 4 milk-first cups.

How many ways can someone choose 4 cups out of 8? That is the combination formula:

C(n, k) = n! / (k! × (n − k)!)

So C(8, 4) = 8! / (4! × 4!) = 40,320 / (24 × 24) = 40,320 / 576 = 70.

Only one of those 70 selections is completely correct. If the taster is purely guessing, her chance of a perfect score is 1/70, about 1.4%.

What about 3 out of 4?

To pick exactly 3 correct milk-first cups and 1 wrong one, she chooses 3 of the 4 right cups and 1 of the 4 wrong cups: C(4, 3) × C(4, 1) = 4 × 4 = 16 ways. Add the 1 perfect selection and you get 17 of 70 selections with at least 3 correct, about 24.3%. Roughly one guesser in four would do that well. That is why Fisher only counted a perfect sort as convincing.

Check: 1 + 16 + 36 + 16 + 1 = 70.

Why not fewer cups?

With 6 cups (3 of each), there are C(6, 3) = 20 possible selections, so a guesser gets a perfect score 1 time in 20, which is 5%. That is a much weaker test. Eight cups is the smallest balanced design that gets the chance of a lucky perfect score below 2%.

Significance in plain words

Fisher’s logic goes like this:

1. Start by assuming the taster cannot tell the difference. Statisticians call this the null hypothesis.
2. Ask how likely the result would be if she were just guessing. That probability is the p-value.
3. If the result would be very unlikely by luck alone, conclude that guessing is not a good explanation.

A perfect score in the 8-cup test has a p-value of 1/70, about 0.014. By the common convention of 0.05 (1 in 20), that is significant. But significance does not prove the taster is infallible; it tells you luck is an unlikely explanation for this result. Fisher himself stressed that a null hypothesis can be disproved by an experiment, but never proved.

The triangle test used in coffee

Coffee tasters and food scientists often use a different design: the triangle test, standardized in sensory science (for example in ISO 4120). You get 3 cups. Two contain the same coffee, one is different. Your job is to pick the odd one out. Coffee taster training, including the Q Grader program, uses this kind of triangulation exercise.

If you are guessing, you will pick the right cup 1 time in 3. One correct answer means very little. What matters is how many rounds you get right. The chance of at least k correct out of n rounds by guessing follows the binomial distribution:

P(X ≥ k) = sum over i from k to n of C(n, i) × (1/3)^i × (2/3)^(n − i)

Worked example: 6 rounds

So in 6 rounds, 5 or more correct is good evidence you really taste a difference. Four out of 6 is encouraging but could still be luck about 1 time in 10.

How many correct answers do you need?

How to run a blind coffee tasting at home

What you need

• Two coffees you want to compare (for example, two single origins, or the same coffee at two roast levels)
• Identical cups, at least 9 small ones
• A kettle, scale and brewer (French press or pour-over)
• A friend to prepare and code the cups, plus sticky notes

Steps

1. Brew each coffee the same way. Same ratio, grind, water and time. For example, 25 g coffee to 400 g water (1:16) per coffee gives enough for 9 small tasting cups of 40 ml each.
2. Let them cool slightly. Around 140–160°F (60–70°C), flavors are easier to tell apart than when piping hot.
3. Have your helper randomize. For each triangle round, they decide which coffee is the odd one, and in which position, by rolling a die or using a random number app. They mark the bottoms of the cups and record the answers.
4. Decide your pass mark first. For 6 rounds, set 5 correct as convincing.
5. Taste in silence. No discussion, no looking at the cup bottoms. Rinse your mouth with water between rounds.
6. Score and reveal. Compare your picks with the answer sheet and check the table above.

Avoid these common mistakes

• Unequal cups: different temperatures or fill levels can give the answer away.
• Visual clues: darker roasts look darker. Use opaque cups or dim the lights.
• The helper’s face: ideally, the person serving does not know the answers either (a double-blind setup).
• Moving the goalposts: deciding afterward that 4 out of 6 was good enough.

Coffees for your experiment

Our Best Sellers Sample Pack is made for this: six 2 oz (about 57 g) packs of 6 Bean Blend, Cowboy Blend, Breakfast Blend, Peru, Mexico and Bali Blue. Each pack is more than enough for one coffee in a 6-round triangle test brewed as above (25 g each). Start with an easy pair, like Bali Blue (medium-dark, dark chocolate and molasses) against Peru (medium, salted caramel and citrus), then challenge yourself with the subtler Peru vs Mexico. Everything is roasted 1–5 days before delivery, so freshness will not muddy the results.

Blind coffee tasting FAQ

Did Muriel Bristol really pass the test?

Accounts from Fisher’s circle, retold in Joan Fisher Box’s biography and later books, say she identified the cups correctly. The exact date and details were not formally recorded.

What is the chance of guessing all 8 cups correctly?

1 in 70, about 1.4%, because there are C(8, 4) = 70 ways to choose 4 cups from 8.

What is a triangle test in coffee?

You taste 3 cups, 2 identical and 1 different, and try to identify the odd one. Guessing gives a 1-in-3 chance per round, so you repeat it several times.

How many rounds should a blind tasting have?

At least 6 triangle rounds. With 6, scoring 5 or more correct would happen by chance less than 2% of the time.

Sources

1. Ronald A. Fisher, The Design of Experiments, Oliver and Boyd, 1935, chapter 2.
2. Joan Fisher Box, R. A. Fisher: The Life of a Scientist, Wiley, 1978.
3. Lady tasting tea, Wikipedia, accessed 2026.
4. Ronald Aylmer Fisher, History of Data Science.
5. ISO 4120:2021 Sensory analysis, methodology, triangle test, International Organization for Standardization.

Facts checked in October 2026.

Keep reading: Single origin vs blend coffee · The best coffee beans in the world · Best selling coffees