Skip to content
Experimento A / B   growth lab

Usability Test Sample Size Calculator: How Many Users?

By the Experimento team | Updated 2026 | method-checked

Enter how often a typical problem shows up per participant and this works out how many people you need to see a target share of the problems, or what share a fixed number of sessions will uncover. The key insight: five users only find about 84% of problems when the average problem hits 31% of people. For a problem that hits 10% of users, five sessions give you a 41% chance of seeing it at all.

How many participants does your usability test need?

Uses the problem discovery model from Nielsen and Landauer: the share of problems found by n participants is 1 − (1 − p)n, where p is the chance that one participant runs into a given problem.

What do you want to work out?
31% is the cross-project average Nielsen and Landauer reported. Use 10% or lower to plan for rarer problems.
Between 1 and 99.9. Finding 100% is never guaranteed by this model.
Groups who use the product differently, such as buyers and sellers. Each needs its own sessions.

How the calculation works

The model treats each problem as something a participant either hits or does not, with the same probability p for everyone. The chance that nobody in n sessions hits it is (1 − p)n, so the chance at least one person does is 1 − (1 − p)n. Averaged across problems of that frequency, the same figure is the share of problems you expect to see. To find the participants needed for a target share D, the calculator solves n = ln(1 − D) / ln(1 − p) and rounds up.

At the 31% average, that gives 84% for five participants, 95% for eight and 98% for ten. At 10% it gives 41%, 57% and 65%. The drop is why the five-users rule holds for the obvious problems and fails for the ones only some people hit.

Where the model breaks down

  • Problems are not equally likely for everyone. A confusing label might trip every novice and no expert. Treat each distinct group of users as its own study, which is what the groups field does.
  • p from a small study runs high. If you estimate the frequency from the same handful of sessions, the problems nobody saw are missing from the count, so p looks bigger than it is and the sample looks sufficient.
  • Your participants may be an unlucky set. The formula gives an expected value. Faulkner's 2003 study found that some random sets of five found only 55% of problems, while sets of 20 never dropped below 95%.
  • It says nothing about measurement. For task times, success rates or SUS scores with a margin of error, you need a statistical sample size, not a discovery one.

For planning the sessions themselves, see our usability testing guide.

// the readout

Get the Experimento newsletter

Independent guides and reviews, straight to your inbox. No spam.

9,400+ growth folks no spam, ever

Confidence 95%. Opt out anytime.