What five people
actually find.
Work out how many participants it takes before another one stops telling you anything new, and get a sentence you can put in the deck when somebody says you should have tested with twenty. This covers qualitative usability testing and nothing else.
Your study
The calculator is prefilled with numbers from the research listed at the bottom of this page. They are averages, taken from other people’s products. Type your own in to get an answer about yours.
What each participant adds
Each bar carries two lines. The dark one is what you find if your low guess is right, the lighter one above it is what you find if your high guess is right. The gap between them is everything you do not know yet. It is widest at the start. Hover a bar to read it, or tab to the chart and use the arrow keys.
5 participants: 84 to 93 percent
Where the number comes from
One participant finds some share of the problems. The second finds some of what the first missed. The third finds some of what those two missed. Every participant has less left to find than the one before, so the curve flattens.
1 - (1 - 31%)^5 = 84% · 1 - (1 - 42%)^5 = 93%
This counts problems found in a usability test. It does not tell you how many of your users run into one of them. It also treats every problem as equally easy to spot. That is not true. A problem only a rare user hits stays rare however long you keep testing.
It does not measure anything. A completion rate, a preference between two concepts, a satisfaction score, a difference between two designs. Every one of those is a number that needs error bars around it, and the error bars come from a different sum. This curve cannot give you one. Asked to, it will hand back a confident percentage that means nothing.
Faulkner tested sixty people on one application. Then she drew smaller sets out of those sixty at random, a hundred times over, to see what a shorter study would have caught.
- 85% on average, and the unluckiest set found 55%.
- 95% on average, and the unluckiest set found 82%.
- 97% on average, and the unluckiest set found 90%.
Where those figures come from
Six sources. Five are behind the numbers on this page, including the two that are sitting in the boxes at the top. The sixth answers a different question, and it is here because people bring that question to this tool. Each one says what it measured and where it stops applying.
- A mathematical model of the finding of usability problems
- The curve this tool runs. They fitted 1 - (1 - p)^n to the problems found across a set of usability projects, where p is the share of problems one participant turns up.
- Published 1993
- It is a model fitted to projects from the early nineties, on software nothing like the thing you are testing. It counts problems found in a usability test. It also assumes every problem is equally easy to spot. That is not true. A problem only a rare user hits stays rare however many sessions you run.
- Why you only need to test with 5 users
- Where the 31 percent detection rate comes from, and the five user rule with it. Nielsen puts the average rate at 31 percent across the studies he analysed.
- Published 2000
- An average across other people’s projects is not your project. The article argues for testing five people again and again through a redesign. So it is an argument for many small rounds of testing. It is often quoted as permission to run one study of five and stop, which is not what it says.
- Refining the test phase of usability evaluation: how many subjects is enough?
- The same curve, fitted to a different set of projects. Virzi put the share one participant finds between 32 and 42 percent, and found that the most serious problems turn up earliest.
- Published 1992
- Three studies, all from one lab, on software from the early nineties. The claim that serious problems surface first has been argued with since, so a small study is not a guarantee that what it missed was minor. The figures here are as reported in summaries of the paper, since it is behind a paywall.
- Eight is not enough
- What happened on a large shopping site. Five participants turned up 35 percent of the problems, and eighteen turned up under half of the six hundred the team eventually counted. Worked backwards, that is a rate near eight percent.
- Published 2001
- One site, and a big complicated one, which is the point of including it. It is the low end of the range and the reason the boxes on this page can be moved. The eight percent is worked out from their reported figures rather than stated by them.
- Beyond the five-user assumption
- Sixty people tested one web application. Sets of five, ten and fifteen were drawn out of those sixty at random, a hundred times over, and each set was scored on the share of the full problem list it found. Five found 85 percent on average. The unluckiest five found 55 percent.
- Published 2003
- One system and one problem list. The hundred percent it scores against is what all sixty participants found between them, not every problem that exists in the product. The figures here are as reported in summaries of the paper, since the paper itself is behind a paywall.
- UX research field guide, surveys chapter
- A minimum of forty participants for a quantitative study, and five to ten for a qualitative one.
- Updated January 2025
- Not used in this tool. Forty is about measuring how many people feel something, to a confidence you can state. This curve counts problems found in a usability test. It cannot tell you how many of your users hit one. That is the most common way this maths gets misused.